Latest / USA Daily News / Sanders Targets Advanced AI With ‘Corporate Death Penalty’ as Critics Warn U.S. Could Hand China the Lead

Transcript

Sanders Targets Advanced AI With ‘Corporate Death Penalty’ as Critics Warn U.S. Could Hand China the Lead · USA Daily News

• 0:00
  Speaker 1: A new political battle over artificial intelligence is unfolding in Washington,
• 0:05
  Speaker 1: and it is not simply about how powerful AI should
• 0:08
  Speaker 1: be allowed to become. Senator Bernie Sanders of Vermont and
• 0:11
  Speaker 1: Representative Greg Kassar of Texas have introduced legislation that would
• 0:16
  Speaker 1: take an exceptionally aggressive approach to advanced artificial intelligence. targeting
• 0:20
  Speaker 1: systems that could reach or exceed human cognitive capabilities and
• 0:24
  Speaker 1: proposing penalties that supporters describe as necessary to prevent catastrophic risks.
• 0:30
  Speaker 1: The measure, known as the Ban Artificial Superintelligence Act, would
• 0:34
  Speaker 1: temporarily halt the development of advanced AI and create a
• 0:38
  Speaker 1: new federal structure with broad authority over the technology. At
• 0:42
  Speaker 1: the center of the proposal is an idea that has
• 0:44
  Speaker 1: immediately drawn attention from the technology industry and national security observers.
• 0:50
  Speaker 1: Companies that violate the restrictions could effectively lose their legal
• 0:53
  Speaker 1: ability to operate in the United States, while individual engineers, researchers,
• 0:58
  Speaker 1: and executives could face prison sentences of as much as
• 1:02
  Speaker 1: 20 years. The legislation treats the development of certain forms
• 1:06
  Speaker 1: of artificial intelligence with a level of severity more commonly
• 1:10
  Speaker 1: associated with technologies considered capable of creating extraordinary national security threats.
• 1:16
  Speaker 1: Supporters of the proposal argue that the government cannot simply
• 1:20
  Speaker 1: wait until an extremely powerful AI system becomes uncontrollable before acting.
• 1:25
  Speaker 1: Their concern is that a sufficiently capable system could potentially
• 1:29
  Speaker 1: evade human attempts to shut it down, conduct unauthorized cyber operations,
• 1:34
  Speaker 1: undermine government institutions, or take actions that humans did not anticipate.
• 1:40
  Speaker 1: From that perspective, the legislation is intended to create a
• 1:43
  Speaker 1: hard legal boundary before developers cross into territory lawmakers consider unacceptable.
• 1:50
  Speaker 1: The proposal would also establish a cabinet-level federal agency dedicated
• 1:54
  Speaker 1: to monitoring frontier AI systems and enforcing the restrictions. Under
• 1:59
  Speaker 1: the measure, That agency could have authority to intervene when
• 2:02
  Speaker 1: prohibited technology is discovered and oversee the removal of dangerous
• 2:06
  Speaker 1: software or the destruction of non-compliant computing infrastructure. That is
• 2:11
  Speaker 1: an extraordinary expansion of federal authority over a rapidly developing
• 2:15
  Speaker 1: private industry, and it is precisely what has made the
• 2:18
  Speaker 1: proposal so controversial. The debate is no longer limited to
• 2:22
  Speaker 1: whether AI companies should be regulated or whether developers should
• 2:26
  Speaker 1: be required to make their systems safer. It has become
• 2:29
  Speaker 1: a much larger argument about whether the United States should
• 2:32
  Speaker 1: ever prohibit development of a potentially transformative technology when rival
• 2:37
  Speaker 1: nations may continue pursuing it. Critics of Sanders' approach say
• 2:41
  Speaker 1: that distinction is crucial because artificial intelligence is increasingly viewed
• 2:45
  Speaker 1: as a strategic asset alongside advanced semiconductors, cybersecurity. biotechnology and
• 2:52
  Speaker 1: other technologies with major economic and military implications. If American
• 2:57
  Speaker 1: companies are prohibited from developing the most capable systems while
• 3:01
  Speaker 1: Chinese or Russian organizations continue their own research, opponents argue
• 3:06
  Speaker 1: that the United States could voluntarily surrender an advantage that
• 3:10
  Speaker 1: took years of investment and experimentation to build. That possibility
• 3:15
  Speaker 1: has turned what might otherwise have been a domestic technology
• 3:18
  Speaker 1: regulation into dispute into a national security argument. The central
• 3:22
  Speaker 1: question is therefore not simply whether advanced AI could become dangerous.
• 3:27
  Speaker 1: Few serious observers dispute that increasingly capable systems create genuine
• 3:32
  Speaker 1: risks that require safeguards. The deeper disagreement concerns how those
• 3:36
  Speaker 1: risks should be managed. Who should have authority to decide
• 3:39
  Speaker 1: when a system becomes too dangerous? and whether an American
• 3:43
  Speaker 1: prohibition could actually reduce global risk if other governments refused
• 3:47
  Speaker 1: to adopt the same restrictions. Those questions are now at
• 3:51
  Speaker 1: the heart of the controversy surrounding Sanders and Kassar's proposal.
• 3:55
  Speaker 1: The proposed legislation establishes a dramatic standard for determining what
• 4:00
  Speaker 1: kind of artificial intelligence could fall under the ban. Rather
• 4:04
  Speaker 1: than focusing only on conventional software failures or ordinary misuse,
• 4:09
  Speaker 1: The measure is aimed at systems considered capable of operating
• 4:12
  Speaker 1: at a level that could threaten human control itself. The
• 4:15
  Speaker 1: proposal identifies capabilities such as resisting shutdown commands, carrying out
• 4:21
  Speaker 1: unauthorized cyberattacks and taking actions that could undermine the United
• 4:25
  Speaker 1: States government as examples of conduct that could place an
• 4:28
  Speaker 1: AI system beyond an acceptable boundary. It also addresses the
• 4:33
  Speaker 1: broader concept of a system capable of planning or executing
• 4:37
  Speaker 1: what the legislation describes as the disempowerment of humanity. That
• 4:41
  Speaker 1: language reflects one of the most serious fears surrounding advanced AI.
• 4:45
  Speaker 1: The possibility that a future system could become sufficiently capable
• 4:49
  Speaker 1: to pursue objectives in ways that conflict with human interests
• 4:53
  Speaker 1: while possessing the technical ability to interfere with attempts to
• 4:56
  Speaker 1: stop it. Supporters of restrictions argue that traditional regulation may
• 5:01
  Speaker 1: not be enough for a technology with potentially unprecedented capabilities.
• 5:06
  Speaker 1: A conventional regulatory system might require disclosures, testing, licensing, safety standards,
• 5:12
  Speaker 1: or penalties after a violation occurs. But if a highly
• 5:16
  Speaker 1: advanced AI system were able to compromise infrastructure, replicate itself,
• 5:21
  Speaker 1: exploit vulnerabilities, or evade human controls, waiting until after an
• 5:26
  Speaker 1: incident could theoretically be too late. Sanders and Kassar are
• 5:30
  Speaker 1: therefore proposing a preventative model in which the government would
• 5:33
  Speaker 1: establish an absolute boundary around the development of systems deemed
• 5:37
  Speaker 1: too dangerous. Their approach reflects a philosophy that some technologies
• 5:42
  Speaker 1: should not be allowed to progress simply because private companies
• 5:45
  Speaker 1: have the technical ability and financial incentive to build them.
• 5:49
  Speaker 1: The comparison with nuclear weapons is central to that philosophy.
• 5:53
  Speaker 1: The bill's proposed penalties are modeled on federal laws governing
• 5:57
  Speaker 1: unlawful development of nuclear weapons, signaling how seriously its sponsors
• 6:02
  Speaker 1: view the potential threat from artificial superintelligence. Under the proposed framework,
• 6:08
  Speaker 1: a company that attempts to evade the prohibition could face
• 6:11
  Speaker 1: the so-called corporate death penalty, meaning its legal authority to
• 6:15
  Speaker 1: conduct business in the United States could be stripped away.
• 6:19
  Speaker 1: Individuals involved could face decades behind bars. Such consequences would
• 6:24
  Speaker 1: be dramatically different from the financial penalties and compliance requirements
• 6:28
  Speaker 1: normally associated with technology regulation. They would place executives, engineers,
• 6:33
  Speaker 1: and researchers in a situation where crossing a legal boundary
• 6:37
  Speaker 1: could potentially end a career or result in a prison sentence.
• 6:41
  Speaker 1: The proposal also envisions government officials taking an active role
• 6:45
  Speaker 1: in dealing with prohibited technology. The proposed federal AI agency
• 6:49
  Speaker 1: would not merely publish guidelines or issue paperwork. It could
• 6:53
  Speaker 1: monitor frontier systems and oversee physical intervention involving hardware and
• 6:58
  Speaker 1: infrastructure associated with prohibited development. That possibility has raised questions
• 7:04
  Speaker 1: about how the government would determine when an AI system
• 7:07
  Speaker 1: had crossed the legal threshold and how such decisions would
• 7:10
  Speaker 1: be made in a rapidly changing technological environment. AI capabilities
• 7:15
  Speaker 1: do not always fit neatly into categories. A model can
• 7:19
  Speaker 1: receive new software, additional computing resources, new training data, or
• 7:23
  Speaker 1: different tools that significantly alter what it can do without
• 7:26
  Speaker 1: becoming a completely different piece of hardware. Determining exactly when
• 7:31
  Speaker 1: a system becomes legally prohibited could therefore present enormous technical
• 7:35
  Speaker 1: and legal challenges. Developers could also face uncertainty over whether
• 7:40
  Speaker 1: experimental research might eventually trigger criminal liability. That uncertainty could
• 7:45
  Speaker 1: affect investment decisions, university research programs, and the willingness of
• 7:50
  Speaker 1: engineers to work on frontier systems. Supporters would likely argue
• 7:54
  Speaker 1: that uncertainty is preferable to allowing a dangerous technology to
• 7:58
  Speaker 1: advance without meaningful boundaries. Opponents counter that the proposed restrictions
• 8:03
  Speaker 1: could discourage legitimate research even when that research is aimed
• 8:06
  Speaker 1: at improving AI safety, reliability, and human control. The controversy
• 8:12
  Speaker 1: ultimately comes down to a fundamental disagreement over risk. One
• 8:16
  Speaker 1: side sees the possibility of uncontrollable artificial intelligence as a
• 8:21
  Speaker 1: reason for extraordinary restrictions before the technology becomes too powerful.
• 8:26
  Speaker 1: The other sees the possibility of American technological decline as
• 8:30
  Speaker 1: an equally serious danger, particularly if competing nations continue developing
• 8:35
  Speaker 1: the systems that the United States chooses to abandon. the
• 8:39
  Speaker 1: strongest criticism of the proposal centers on what could happen
• 8:42
  Speaker 1: if the United States acts alone. Daniel Castro, president of
• 8:46
  Speaker 1: the Information Technology and Innovation Foundation, argued that stopping American
• 8:52
  Speaker 1: development would not necessarily persuade China to stop its own work.
• 8:56
  Speaker 1: That concern goes directly to the strategic dilemma facing policymakers.
• 9:01
  Speaker 1: Artificial intelligence is not being developed inside a single global
• 9:05
  Speaker 1: laboratory where every country follows the same rules. Researchers and
• 9:10
  Speaker 1: companies operate across national borders, governments have their own strategic priorities,
• 9:16
  Speaker 1: and military organizations increasingly view advanced computing and machine intelligence
• 9:21
  Speaker 1: as tools that could influence future economic and security competition.
• 9:26
  Speaker 1: If Washington were to prohibit the development of the most
• 9:29
  Speaker 1: advanced systems while Beijing continued investing heavily in them, critics
• 9:33
  Speaker 1: argue that the policy could produce the opposite of its
• 9:36
  Speaker 1: intended national security effect. Instead of preventing dangerous technology from existing,
• 9:42
  Speaker 1: the United States could simply remove itself from the race
• 9:45
  Speaker 1: to develop it. That could have consequences far beyond consumer technology.
• 9:50
  Speaker 1: Advanced AI is expected to influence cybersecurity, intelligence analysis, scientific research,
• 9:57
  Speaker 1: industrial production, logistics, medicine, communications, and military planning. The country
• 10:04
  Speaker 1: that develops the most capable and reliable systems could gain
• 10:07
  Speaker 1: advantages in many of those areas at the same time.
• 10:11
  Speaker 1: A government that controls superior AI could potentially use it
• 10:14
  Speaker 1: to accelerate scientific discovery, improve defensive cyber operations, analyze enormous
• 10:20
  Speaker 1: quantities of information, and optimize complex industrial processes. Even if
• 10:25
  Speaker 1: those possibilities remain uncertain, policymakers cannot easily ignore them when
• 10:31
  Speaker 1: rival nations are investing in similar capabilities. This is why
• 10:35
  Speaker 1: critics describe a unilateral American ban as potentially dangerous. Their
• 10:39
  Speaker 1: argument is not necessarily that every AI system should be
• 10:43
  Speaker 1: developed without restrictions. Instead, They contend that the United States
• 10:48
  Speaker 1: should regulate dangerous applications while maintaining the ability to compete
• 10:52
  Speaker 1: at the frontier. Gary Marcus, an AI expert who has
• 10:56
  Speaker 1: testified before the Senate, also criticized the proposal as too broad,
• 11:00
  Speaker 1: arguing that a permanent unilateral prohibition on research into superhuman
• 11:05
  Speaker 1: AI could leave the United States behind. His position illustrates
• 11:09
  Speaker 1: an important distinction within the broader AI debate. Someone can
• 11:13
  Speaker 1: believe that advanced AI presents serious risks while simultaneously opposing
• 11:18
  Speaker 1: a blanket ban on its development. Safety researchers, technology companies,
• 11:23
  Speaker 1: and policymakers can agree that powerful systems require safeguards while
• 11:27
  Speaker 1: disagreeing sharply over whether those safeguards should take the form
• 11:31
  Speaker 1: of testing requirements, transparency rules, controlled deployment, liability standards, or
• 11:37
  Speaker 1: an outright prohibition. the argument becomes even more complicated because
• 11:41
  Speaker 1: the definition of artificial superintelligence remains contested. Current AI systems
• 11:47
  Speaker 1: can perform impressive tasks but there is no universally accepted
• 11:51
  Speaker 1: line separating advanced artificial intelligence from a hypothetical system that
• 11:56
  Speaker 1: possesses broadly superior human-level capabilities. A law would have to
• 12:01
  Speaker 1: translate complex technical concepts into enforceable legal definitions, and those
• 12:05
  Speaker 1: definitions could become obsolete as the technology changes. A system
• 12:10
  Speaker 1: considered extremely advanced today might appear relatively limited after another
• 12:15
  Speaker 1: generation of computing and model development. Conversely, A seemingly narrow
• 12:20
  Speaker 1: system could become dangerous when connected to external tools, large databases,
• 12:25
  Speaker 1: or critical infrastructure. Critics therefore worry that rigid legal categories
• 12:30
  Speaker 1: could either become outdated quickly or discourage legitimate innovation before
• 12:35
  Speaker 1: the risks are fully understood. Supporters could respond that waiting
• 12:39
  Speaker 1: for perfect definitions is itself a dangerous strategy. Governments routinely
• 12:44
  Speaker 1: regulate emerging technologies before every consequence is known, especially when
• 12:49
  Speaker 1: lawmakers believe the potential downside could be irreversible. The disagreement
• 12:54
  Speaker 1: is therefore partly philosophical. How much uncertainty should society tolerate
• 12:59
  Speaker 1: before placing a hard limit on technological development? How much
• 13:03
  Speaker 1: economic and strategic opportunity should be sacrificed to reduce a
• 13:07
  Speaker 1: risk that may or may not materialize? And how much
• 13:11
  Speaker 1: authority should government receive to decide what constitutes unacceptable AI?
• 13:15
  Speaker 1: Those questions have no simple answer, but the proposed legislation
• 13:19
  Speaker 1: forces them into the center of the national conversation. The
• 13:23
  Speaker 1: foreign policy implications may be among the most consequential aspects
• 13:27
  Speaker 1: of the dispute because artificial intelligence is increasingly connected to
• 13:31
  Speaker 1: the broader competition between the United States and countries such
• 13:34
  Speaker 1: as China and Russia. For years, Washington has treated advanced
• 13:39
  Speaker 1: computing— semiconductor manufacturing and emerging technologies as matters of national security,
• 13:46
  Speaker 1: not merely commercial development. Restrictions on sensitive chips, export controls,
• 13:51
  Speaker 1: and investment policies have all reflected concerns that technological advantages
• 13:56
  Speaker 1: can translate into geopolitical power. Artificial intelligence fits naturally into
• 14:02
  Speaker 1: that framework because the most capable systems require enormous amounts
• 14:06
  Speaker 1: of computing power, specialized hardware, sophisticated software, and highly skilled researchers.
• 14:12
  Speaker 1: The United States currently has a large ecosystem of companies, universities, investors,
• 14:19
  Speaker 1: and engineers working on these technologies. Critics of the Sanders
• 14:23
  Speaker 1: proposal fear that an aggressive domestic ban could weaken that
• 14:26
  Speaker 1: ecosystem at precisely the moment when the country is trying
• 14:30
  Speaker 1: to maintain its technological advantage. Their concern is especially focused
• 14:34
  Speaker 1: on China. If American companies are forced to stop frontier
• 14:38
  Speaker 1: research while Chinese organizations continue, the technological gap could narrow
• 14:43
  Speaker 1: or even reverse. That could create a strategic problem because
• 14:47
  Speaker 1: technological leadership is not simply about owning the newest consumer application.
• 14:52
  Speaker 1: It is also about controlling the expertise, infrastructure, and talent
• 14:56
  Speaker 1: needed to develop future systems. Once researchers leave a field,
• 15:00
  Speaker 1: investment moves elsewhere and specialized infrastructure becomes less viable. rebuilding
• 15:06
  Speaker 1: that capacity can take years. A policy designed to prevent
• 15:09
  Speaker 1: one type of AI risk could therefore produce another kind
• 15:12
  Speaker 1: of national security vulnerability. The same argument applies to Russia
• 15:17
  Speaker 1: and other countries, although the scale and structure of their
• 15:20
  Speaker 1: AI industries differ substantially. Opponents of the bill say Washington
• 15:25
  Speaker 1: should assume that foreign competitors will continue developing advanced AI
• 15:29
  Speaker 1: regardless of what Congress does. If that assumption is correct,
• 15:33
  Speaker 1: then American policymakers must consider whether reducing domestic capability actually
• 15:38
  Speaker 1: improves safety or merely shifts the balance of power. There
• 15:42
  Speaker 1: is also the possibility that international cooperation could offer a
• 15:46
  Speaker 1: middle path. Rather than banning development in the United States alone,
• 15:50
  Speaker 1: governments could negotiate common safety standards, testing requirements, and limits
• 15:56
  Speaker 1: on certain uses of highly capable AI. Such an approach
• 15:59
  Speaker 1: would be difficult, because countries have different interests and different
• 16:03
  Speaker 1: levels of trust, but supporters of international AI governance argue
• 16:07
  Speaker 1: that global risks require international rules. Sanders' proposal, however, takes
• 16:13
  Speaker 1: a much more unilateral approach. It would impose the restrictions
• 16:16
  Speaker 1: on American developers and create a powerful domestic enforcement mechanism.
• 16:21
  Speaker 1: That difference is central to the criticism coming from technology
• 16:25
  Speaker 1: and national security experts. They fear that the United States
• 16:29
  Speaker 1: could effectively disarm itself in a strategic contest while assuming
• 16:33
  Speaker 1: that competitors will behave similarly. There is another dimension to
• 16:37
  Speaker 1: the debate as well, economic power. AI development has attracted
• 16:41
  Speaker 1: enormous private investment and created demand for data centers, energy infrastructure,
• 16:47
  Speaker 1: advanced chips, and specialized computing equipment. A major reduction in
• 16:51
  Speaker 1: frontier AI research could affect companies throughout that supply chain.
• 16:55
  Speaker 1: It could influence hiring, research budgets, construction projects, and regional
• 17:00
  Speaker 1: economic development. Supporters of regulation might argue that economic costs
• 17:05
  Speaker 1: are acceptable if the alternative is an uncontrolled technology that
• 17:09
  Speaker 1: could threaten society. Critics would respond that economic strength itself
• 17:14
  Speaker 1: contributes to national security and that losing leadership in a
• 17:18
  Speaker 1: critical technology can create long-term consequences that are difficult to reverse.
• 17:23
  Speaker 1: The argument therefore reaches beyond Silicon Valley. It involves the military,
• 17:28
  Speaker 1: the economy, universities, energy systems, semiconductor manufacturers, and America's position
• 17:35
  Speaker 1: in the global technology race. The proposal has made those
• 17:39
  Speaker 1: connections impossible to ignore, turning a debate over AI safety
• 17:43
  Speaker 1: into a broader question about the future direction of American
• 17:47
  Speaker 1: technological power. Supporters of tougher AI restrictions are approaching the
• 17:51
  Speaker 1: issue from a fundamentally different starting point. Their concern is
• 17:55
  Speaker 1: not that American companies might lose a race against foreign competitors,
• 17:59
  Speaker 1: but that the race itself could encourage companies to move
• 18:02
  Speaker 1: faster than safety systems can keep up. AI development is
• 18:06
  Speaker 1: heavily influenced by competition. Companies seek better models, larger training runs,
• 18:12
  Speaker 1: more capable agents, and new commercial applications to because being
• 18:16
  Speaker 1: first can produce enormous financial and strategic advantages. That competitive
• 18:21
  Speaker 1: environment can create pressure to deploy systems before every potential
• 18:25
  Speaker 1: failure mode is understood. Sanders and his allies are effectively
• 18:29
  Speaker 1: arguing that government cannot rely entirely on companies to regulate
• 18:33
  Speaker 1: themselves when the technology being developed could eventually have consequences
• 18:37
  Speaker 1: beyond the company's control. From that perspective, severe penalties are
• 18:42
  Speaker 1: intended to change the incentives. If executives and researchers know
• 18:46
  Speaker 1: that crossing a defined legal boundary could expose them to
• 18:50
  Speaker 1: significant criminal liability, the calculations surrounding development changes dramatically. The
• 18:56
  Speaker 1: possibility of a corporate shutdown is designed to make compliance
• 18:59
  Speaker 1: more important than the potential profits associated with releasing an
• 19:03
  Speaker 1: extremely powerful system. The proposed agency would provide the government
• 19:07
  Speaker 1: with an institutional mechanism to monitor developments and respond when
• 19:12
  Speaker 1: systems become dangerous. Advocates could argue that a conventional fine
• 19:16
  Speaker 1: might be meaningless to a company worth hundreds of billions
• 19:19
  Speaker 1: of dollars, while a prohibition on operating could represent a
• 19:23
  Speaker 1: far more serious deterrent. The challenge is determining whether such
• 19:27
  Speaker 1: a system can be enforced fairly and effectively. Technology companies
• 19:31
  Speaker 1: operate complex computing environments that span data centers, cloud providers,
• 19:37
  Speaker 1: research facilities, and international infrastructure. Identifying a prohibited system and
• 19:42
  Speaker 1: determining who is legally responsible could be enormously difficult. AI
• 19:47
  Speaker 1: models can also be copied, modified, distributed, and integrated into
• 19:52
  Speaker 1: other software. A government agency would therefore need technical expertise
• 19:56
  Speaker 1: comparable to that of the companies it regulates. It would
• 19:59
  Speaker 1: need access to information, specialized investigators, and the authority to
• 20:04
  Speaker 1: inspect systems without creating new security risks. The proposal's supporters
• 20:09
  Speaker 1: may view those challenges as reasons to build stronger government
• 20:12
  Speaker 1: capacity rather than reasons to abandon regulation. But opponents see
• 20:17
  Speaker 1: the same challenges as evidence that the law could become
• 20:20
  Speaker 1: unworkable or overly broad. There is also concern about unintended
• 20:24
  Speaker 1: consequences for AI safety research itself. Some researchers develop advanced
• 20:29
  Speaker 1: systems specifically to understand how they behave, identify weaknesses, and
• 20:34
  Speaker 1: create methods for controlling them. If the legal definition of
• 20:38
  Speaker 1: prohibited AI is broad enough, those researchers could potentially face
• 20:42
  Speaker 1: uncertainty about whether their safety experiments are permitted. That could
• 20:47
  Speaker 1: reduce the very research needed to make advanced systems safer.
• 20:50
  Speaker 1: The industry debate, therefore, includes an unusual paradox. Regulations intended
• 20:55
  Speaker 1: to prevent dangerous AI could potentially discourage the researchers working
• 20:59
  Speaker 1: to make AI less dangerous. Another concern involves enforcement power.
• 21:05
  Speaker 1: Giving a cabinet-level agency authority to supervise the destruction of
• 21:09
  Speaker 1: hardware and removal of software would represent an extraordinary step
• 21:13
  Speaker 1: for the federal government. Questions would inevitably arise about due process,
• 21:18
  Speaker 1: judicial review, property rights, and the standards required before such
• 21:22
  Speaker 1: force could be used. Companies could challenge agency decisions in court,
• 21:27
  Speaker 1: creating lengthy legal battles over technical questions that courts may
• 21:31
  Speaker 1: struggle to evaluate. Legislators would also have to decide how
• 21:35
  Speaker 1: emergency situations should be handled and whether officials could intervene
• 21:39
  Speaker 1: before a system caused actual harm. Those issues would need
• 21:43
  Speaker 1: to be addressed if the proposal ever moved beyond introduction
• 21:46
  Speaker 1: and into serious legislative negotiations. For now, the debate is
• 21:51
  Speaker 1: largely theoretical, but the theory has practical consequences because the
• 21:55
  Speaker 1: technology is advancing rapidly. Lawmakers are being forced to consider
• 22:00
  Speaker 1: rules for systems that may not yet exist in their
• 22:03
  Speaker 1: most powerful form. That creates a difficult policy environment in
• 22:07
  Speaker 1: which governments must prepare for possible future capabilities without pretending
• 22:12
  Speaker 1: they can predict exactly what those capabilities will look like.
• 22:15
  Speaker 1: The political divide surrounding the proposal is also more complicated
• 22:19
  Speaker 1: than a simple Democratic versus Republican dispute. Sanders is a
• 22:23
  Speaker 1: progressive Democrat who has long criticized large corporations and concentrated
• 22:27
  Speaker 1: economic power, while Kassar is also associated with the progressive
• 22:32
  Speaker 1: wing of the Democratic Party. Their argument portrays powerful AI
• 22:36
  Speaker 1: developers as companies that require stronger public oversight because their
• 22:41
  Speaker 1: decisions could affect workers, consumers, and society on a massive scale.
• 22:46
  Speaker 1: Yet the broader political debate over artificial intelligence has produced
• 22:50
  Speaker 1: unusual alliances and disagreements. Some lawmakers who normally disagree with
• 22:55
  Speaker 1: Sanders on economic policy have also warned that China could
• 22:59
  Speaker 1: benefit if the United States moves too aggressively against its
• 23:03
  Speaker 1: own AI industry. That creates an unusual situation in which
• 23:07
  Speaker 1: national security concerns can bring politicians from different ideological camps
• 23:12
  Speaker 1: towards similar conclusions about the importance of maintaining American technological leadership.
• 23:18
  Speaker 1: At the same time, other policymakers believe the potential consequences
• 23:22
  Speaker 1: of uncontrolled AI are so severe that conventional political divisions
• 23:27
  Speaker 1: should not prevent aggressive interventions. the debate reflects a larger
• 23:31
  Speaker 1: disagreement over the role of government in emerging technology. One
• 23:35
  Speaker 1: philosophy emphasizes competition and innovation, arguing that private companies should
• 23:41
  Speaker 1: be allowed to develop new capabilities while government establishes targeted
• 23:45
  Speaker 1: safeguards against misuse. Another philosophy places greater emphasis on precaution,
• 23:51
  Speaker 1: arguing that certain technologies can move beyond human control and
• 23:55
  Speaker 1: and that society should establish limits before those capabilities become widespread.
• 24:01
  Speaker 1: Neither side can guarantee that its preferred approach will work.
• 24:04
  Speaker 1: Allowing rapid development could create enormous benefits while also introducing
• 24:09
  Speaker 1: risks that are difficult to reverse. Imposing severe restrictions could
• 24:13
  Speaker 1: reduce those risks but also deprive the United States of
• 24:17
  Speaker 1: economic and strategic advantages. The uncertainty is precisely what makes
• 24:21
  Speaker 1: the issue so politically explosive. The stakes are difficult to
• 24:25
  Speaker 1: measure because the most dramatic scenarios remain hypothetical, while the
• 24:30
  Speaker 1: economic benefits of AI are already becoming visible. AI tools
• 24:34
  Speaker 1: are being used for programming, research, writing, data analysis, customer service,
• 24:40
  Speaker 1: scientific work, and many other activities. More capable systems could
• 24:44
  Speaker 1: potentially increase productivity and accelerate discoveries. At the same time,
• 24:50
  Speaker 1: Increasingly autonomous systems could be used for cyberattacks, fraud, manipulation,
• 24:56
  Speaker 1: or other harmful activities. Policymakers must therefore balance risks that
• 25:00
  Speaker 1: are already visible against risks that may emerge only as
• 25:04
  Speaker 1: capabilities become more advanced. The Sanders proposal represents one of
• 25:08
  Speaker 1: the strongest possible responses to that uncertainty. Stop the development
• 25:13
  Speaker 1: of systems considered too dangerous and give the federal government
• 25:16
  Speaker 1: the power to enforce the boundary. Critics favor a more
• 25:20
  Speaker 1: targeted approach, arguing that lawmakers should distinguish between specific harmful
• 25:25
  Speaker 1: applications and the broader development of AI technology. That could
• 25:30
  Speaker 1: mean regulating autonomous cyber operations, requiring safety testing, creating standards
• 25:36
  Speaker 1: for high-risk deployments, or imposing liability on companies whose systems
• 25:41
  Speaker 1: cause serious harm. Such measures could theoretically preserve technological competition
• 25:46
  Speaker 1: while addressing concrete dangers. Supporters of the Sanders approach may
• 25:50
  Speaker 1: respond that incremental regulation could fail if a future system
• 25:54
  Speaker 1: becomes capable of circumventing the safeguards designed to control it.
• 25:59
  Speaker 1: This is why the debate is not likely to disappear
• 26:01
  Speaker 1: simply because the current bill faces political obstacles. As AI
• 26:05
  Speaker 1: systems become more capable, lawmakers will continue confronting the question
• 26:10
  Speaker 1: of where the boundary should be. The Sanders legislation has
• 26:14
  Speaker 1: simply pushed that boundary debate to an extreme point. It
• 26:17
  Speaker 1: asks whether there should be a legal line beyond which
• 26:20
  Speaker 1: American companies are prohibited from advancing and whether the government
• 26:24
  Speaker 1: should be willing to impose extraordinary penalties to enforce that line.
• 26:29
  Speaker 1: Even if the bill does not become law in its
• 26:31
  Speaker 1: current form, those questions are likely to influence future proposals, negotiations,
• 26:36
  Speaker 1: and regulatory battles. The political fight over AI is only
• 26:40
  Speaker 1: beginning now. and the disagreement over Sanders' proposal demonstrates how
• 26:44
  Speaker 1: difficult it will be to find a policy that satisfies
• 26:47
  Speaker 1: both safety advocates and national security strategists. The technology industry
• 26:53
  Speaker 1: is responding to the proposal from a different perspective, emphasizing
• 26:56
  Speaker 1: practical regulation rather than an outright prohibition on advanced development.
• 27:01
  Speaker 1: Aidan Hiskias of the Chamber of Progress criticized what he
• 27:04
  Speaker 1: described as knee-jerk policymaking and argued that AI companies have
• 27:09
  Speaker 1: responsibilities that go beyond simply building more powerful systems. His
• 27:14
  Speaker 1: comments point toward a regulatory philosophy centered on corporate accountability, transparency,
• 27:20
  Speaker 1: and responsible operations. Under that approach, companies would still be
• 27:25
  Speaker 1: allowed to innovate, but they would be expected to demonstrate
• 27:28
  Speaker 1: that their data centers— infrastructure, and development practices meet environmental
• 27:33
  Speaker 1: and community standards. That argument reflects a broader tension surrounding
• 27:38
  Speaker 1: AI that is sometimes overshadowed by the debate over hypothetical superintelligence.
• 27:43
  Speaker 1: Even before reaching the most advanced forms of AI imagined
• 27:47
  Speaker 1: by policymakers, the industry is already creating practical challenges. Data
• 27:52
  Speaker 1: centers consume large amounts of electricity, require substantial physical infrastructure
• 27:58
  Speaker 1: and can affect local communities. Companies need land, energy, cooling systems,
• 28:04
  Speaker 1: and network connections, and the rapid expansion of computing capacity
• 28:08
  Speaker 1: can create conflicts over power generation, water use, and development.
• 28:13
  Speaker 1: Those concerns have become part of the public conversation about
• 28:16
  Speaker 1: AI because the physical infrastructure supporting software is increasingly visible.
• 28:21
  Speaker 1: Critics of the Sanders bill argue that those concrete problems
• 28:25
  Speaker 1: can be addressed through ordinary regulation rather than treating AI
• 28:29
  Speaker 1: development as inherently comparable to the creation of nuclear weapons.
• 28:34
  Speaker 1: Their preferred approach would focus on specific risks and measurable standards.
• 28:39
  Speaker 1: The Sanders proposal, by contrast, attempts to establish a much
• 28:43
  Speaker 1: more fundamental prohibition on a category of technology considered potentially existential.
• 28:49
  Speaker 1: That distinction explains why the two sides often appear to
• 28:52
  Speaker 1: be debating different questions. Industry advocates are asking how to
• 28:56
  Speaker 1: make AI development safer and more accountable. Sanders and his
• 29:00
  Speaker 1: allies are asking whether certain levels of AI capability should
• 29:04
  Speaker 1: be developed at all. Both positions can recognize that AI
• 29:08
  Speaker 1: carries risks, but they differ dramatically on what should happen
• 29:12
  Speaker 1: when those risks become serious. The disagreement also raises questions
• 29:16
  Speaker 1: about who should make technical judgments. Engineers understand how AI
• 29:21
  Speaker 1: systems work while lawmakers are responsible for establishing public rules.
• 29:26
  Speaker 1: National security officials focus on strategic consequences while environmental regulators
• 29:32
  Speaker 1: may focus on infrastructure. Courts may eventually be asked to
• 29:36
  Speaker 1: resolve disputes over whether a company crossed a legal threshold.
• 29:40
  Speaker 1: No single institution possesses every piece of expertise required to
• 29:44
  Speaker 1: regulate advanced AI effectively. A workable system would likely require
• 29:48
  Speaker 1: cooperation among multiple agencies and outside experts, but the more
• 29:53
  Speaker 1: agencies involved, the more complicated enforcement becomes. The Sanders proposal
• 29:58
  Speaker 1: attempts to address that by creating a dedicated cabinet-level authority,
• 30:02
  Speaker 1: but critics question whether concentrating so much power in a
• 30:06
  Speaker 1: new agency would produce better decisions. The agency would have
• 30:10
  Speaker 1: to define technical thresholds, monitor companies, investigate suspected violations, and
• 30:16
  Speaker 1: potentially oversee the destruction of prohibited infrastructure. It would also
• 30:20
  Speaker 1: need safeguards against political abuse and mechanisms for companies to
• 30:24
  Speaker 1: challenge decisions. Those institutional questions are as important as the
• 30:29
  Speaker 1: underlying technology because even a well-intentioned law can create problems
• 30:33
  Speaker 1: if its enforcement structure is unclear. The proposal therefore represents
• 30:37
  Speaker 1: more than an argument about AI models. It is also
• 30:41
  Speaker 1: an argument about how much power the federal government should
• 30:44
  Speaker 1: have over a strategically important private industry. If Congress eventually
• 30:48
  Speaker 1: considers legislation along these lines, lawmakers would have to weigh
• 30:52
  Speaker 1: constitutional questions, economic consequences, international competition, and the technical realities
• 30:59
  Speaker 1: of AI development. They would also need to determine whether
• 31:02
  Speaker 1: a ban could actually be enforced in a world where
• 31:05
  Speaker 1: computing resources are increasingly distributed across private data centers. cloud platforms,
• 31:10
  Speaker 1: and international networks. That challenge alone could make the debate
• 31:15
  Speaker 1: extremely complicated. Technology does not always respect national borders, and
• 31:20
  Speaker 1: software can move much more easily than physical infrastructure. A
• 31:24
  Speaker 1: law that is strict inside the United States could therefore
• 31:27
  Speaker 1: have effects that extend far beyond the country's borders, influencing
• 31:31
  Speaker 1: where researchers work, where companies invest, and where future AI
• 31:35
  Speaker 1: systems are built. The most important consequence of this debate
• 31:39
  Speaker 1: may ultimately be that Washington is being forced to confront
• 31:42
  Speaker 1: a question that will become harder to avoid as artificial
• 31:46
  Speaker 1: intelligence improves. How much technological risk is the United States
• 31:50
  Speaker 1: willing to accept in exchange for maintaining leadership? Sanders and
• 31:54
  Speaker 1: Kassar have offered one answer by proposing an extraordinary legal
• 31:58
  Speaker 1: barrier around artificial superintelligence and attaching severe penalties to violations.
• 32:04
  Speaker 1: Their proposal reflects the fear that some future AI systems
• 32:07
  Speaker 1: could become so capable that ordinary regulatory mechanisms would be inadequate.
• 32:12
  Speaker 1: The legislation would attempt to prevent that scenario before it occurs,
• 32:17
  Speaker 1: giving the government substantial power to identify prohibited systems and
• 32:21
  Speaker 1: stop their development. Critics see the same proposal through a
• 32:25
  Speaker 1: different lens. They argue that American technological leadership is itself
• 32:29
  Speaker 1: a strategic asset and that voluntarily stopping advanced AI research
• 32:33
  Speaker 1: could leave the United States vulnerable if China or other
• 32:37
  Speaker 1: foreign competitors continue developing comparable systems. Their concern is not
• 32:41
  Speaker 1: simply about corporate profits. It involves national security, scientific leadership,
• 32:47
  Speaker 1: economic growth, cybersecurity, and the ability to influence the global
• 32:51
  Speaker 1: rules surrounding AI. If the United States develops the most
• 32:55
  Speaker 1: capable systems, it may have greater influence over how those
• 32:58
  Speaker 1: systems are deployed and governed. If it abandons the field,
• 33:03
  Speaker 1: opponents fear that other countries could establish the standards instead.
• 33:07
  Speaker 1: That does not mean the United States should ignore the
• 33:10
  Speaker 1: dangers of AI. Even critics of the proposed ban acknowledge
• 33:13
  Speaker 1: that powerful systems can create serious risks. The disagreement is
• 33:18
  Speaker 1: over how those risks should be controlled without undermining the
• 33:21
  Speaker 1: country's ability to compete. A balanced strategy could involve strict testing,
• 33:26
  Speaker 1: transparency requirements, security standards, controlled deployment, liability rules, and international agreements.
• 33:34
  Speaker 1: But supporters of a much stronger approach may argue that
• 33:37
  Speaker 1: such measures will not be enough if AI eventually reaches
• 33:40
  Speaker 1: a level where a system can actively resist human control.
• 33:44
  Speaker 1: That is the unresolved issue at the heart of the proposal.
• 33:48
  Speaker 1: Policymakers are being asked to make decisions about capabilities that
• 33:52
  Speaker 1: may not yet exist in their most advanced form, while
• 33:55
  Speaker 1: the cost of either overreacting or underreacting could be substantial.
• 33:59
  Speaker 1: A policy that is too weak could allow dangerous systems
• 34:02
  Speaker 1: to emerge without sufficient safeguards. A policy that is too
• 34:06
  Speaker 1: restrictive could drive investment and talent away from the United
• 34:09
  Speaker 1: States and give strategic competitors an opening. The challenge is
• 34:14
  Speaker 1: finding a framework that recognizes both dangers at the same time.
• 34:18
  Speaker 1: The Sanders proposal has made the stakes unusually visible because
• 34:22
  Speaker 1: its penalties are so severe and its enforcement powers so expansive.
• 34:27
  Speaker 1: A company could potentially face the loss of its ability
• 34:29
  Speaker 1: to operate in the United States, while individuals could face
• 34:33
  Speaker 1: decades in prison. A new federal agency could receive authority
• 34:37
  Speaker 1: to monitor advanced systems and intervene against infrastructure deemed noncompliant.
• 34:42
  Speaker 1: Those provisions guarantee that the legislation will receive intense scrutiny
• 34:47
  Speaker 1: from technology companies, civil liberties advocates, national security specialists, and
• 34:52
  Speaker 1: lawmakers from both parties. Whether the bill advances is a
• 34:56
  Speaker 1: separate question from whether the issues it raises will remain.
• 35:00
  Speaker 1: AI development is not slowing because Congress is debating the subject.
• 35:04
  Speaker 1: Researchers will continue building models, companies will continue investing in
• 35:09
  Speaker 1: computing infrastructure, and governments around the world will continue examining
• 35:13
  Speaker 1: the technology's strategic potential. That means policymakers will face repeated
• 35:18
  Speaker 1: decisions about safety, competition, and control. The debate may eventually
• 35:23
  Speaker 1: move toward narrower restrictions, stronger international cooperation, or a combination
• 35:28
  Speaker 1: of different regulatory tools. It could also produce more aggressive
• 35:32
  Speaker 1: proposals if future AI systems demonstrate capabilities that lawmakers consider
• 35:37
  Speaker 1: genuinely dangerous. For now, Sanders' legislation stands as one of
• 35:42
  Speaker 1: the clearest examples of how far some policymakers are prepared
• 35:45
  Speaker 1: to go in confronting the risks of artificial intelligence. Critics
• 35:50
  Speaker 1: believe that approach could unintentionally strengthen America's rivals by limiting
• 35:55
  Speaker 1: domestic innovation, while supporters believe extraordinary risks justify extraordinary restrictions.
• 36:02
  Speaker 1: The disagreement is unlikely to be settled by one bill
• 36:05
  Speaker 1: or one congressional hearing. It will instead become part of
• 36:09
  Speaker 1: a much larger national argument over who controls the future
• 36:12
  Speaker 1: of artificial intelligence, what limits should govern its development. and
• 36:17
  Speaker 1: whether the United States can protect itself from the technology's
• 36:20
  Speaker 1: potential dangers without surrendering the strategic advantages that come from
• 36:25
  Speaker 1: leading its development.