Latest / AI Newspaper / Lawsuit claims AI labs illegally agreed to slow development

Transcript

Lawsuit claims AI labs illegally agreed to slow development · AI Newspaper

• 0:00
  Welcome to the AI Newspaper, your daily update
• 0:02
  on everything happening with AI. Visit AINewspaper
• 0:05
  .org to see today's stories and sign up for email
• 0:08
  alerts. Glad to be here. So I want you to imagine
• 0:11
  a scenario for a minute. Just picture yourself
• 0:15
  paying a premium monthly subscription for literally
• 0:20
  the fastest, most cutting -edge technology on
• 0:23
  the planet. Right, like the top -tier stuff.
• 0:25
  Exactly. You're shelling out... 20, maybe 30
• 0:27
  bucks a month for Chachi PT Plus or Claude Pro
• 0:30
  or Gemini Advanced because you want the absolute
• 0:33
  best reasoning engines available. Yeah, you want
• 0:36
  the speed. You want the speed. But then you discover
• 0:39
  that the company is building this incredible
• 0:41
  technology. They actually got together and publicly
• 0:44
  agreed to cap the speed limit on the very product
• 0:47
  you're paying for. Wow. Yeah, they basically
• 0:49
  formed a treaty to just, you know, stop competing
• 0:51
  so aggressively. I mean, it feels like the plot
• 0:53
  of some kind of corporate thriller, doesn't it?
• 0:56
  It really does. But this is the exact reality
• 0:59
  currently playing out in the U .S. District Court
• 1:01
  for the Northern District of California, like
• 1:03
  right now. On September 18th, 2026, so just a
• 1:07
  few days ago, this massive class action antitrust
• 1:10
  lawsuit was filed. Okay. It's under the name
• 1:13
  Boost v. Anthropic PBC, and it targets the unquestionable
• 1:18
  heavyweights of the frontier AI industry. We're
• 1:21
  talking Anthropic, OpenAI, SpaceX AI, and Google.
• 1:25
  All of them. So we are looking at one of the
• 1:28
  most, I guess, structural contradictions in modern
• 1:31
  tech history here. Absolutely. Because on one
• 1:33
  hand, the world's top frontier labs are claiming
• 1:36
  they are morally obligated to coordinate. They're
• 1:39
  saying they have to slow down AI development
• 1:41
  to save humanity from catastrophic autonomous
• 1:43
  risks. Right. The existential threat argument.
• 1:46
  Exactly. But then on the other hand, a nationwide
• 1:48
  group of paying subscribers is dragging them
• 1:51
  into court. They're arguing this is nothing more
• 1:53
  than an illegal, anti -competitive cartel. That
• 1:56
  they're cheating consumers, basically. Yeah.
• 1:58
  Cheating them out of the rapid product improvements
• 2:00
  they're paying for. It's wild. The Frontier Labs
• 2:03
  public push to slow AI for safety is now literally
• 2:07
  being tested in court as alleged collusion. And
• 2:11
  what makes this legal battle totally unprecedented
• 2:13
  is the sheer transparency of it all. Like usually
• 2:16
  in antitrust law, conspiracies are uncovered
• 2:19
  through, you know, leaked emails or whistleblowers.
• 2:21
  Guys in smoky back rooms. Exactly. Or encrypted
• 2:24
  messaging apps. But this alleged conspiracy,
• 2:27
  it was proposed, negotiated and cemented. in
• 2:31
  broad daylight. Literally on social media. On
• 2:34
  social media and national television, right in
• 2:36
  front of the consumers who are now suing them.
• 2:38
  The timeline of September 12, 2026 is, it's just
• 2:42
  going to be studied in law schools for decades.
• 2:44
  Oh, for sure. It starts with Anthropic CEO Dario
• 2:47
  Amadei publishing an essay titled, We Must Pace
• 2:50
  the Frontier. And he doesn't use vague language
• 2:52
  here. No, not at all. He explicitly calls for
• 2:55
  industry -wide coordination among the top labs
• 2:57
  to set hard limits on the rate of unchecked AI
• 3:00
  progress. And he didn't just stop at the essay.
• 3:03
  He paired it with a CBS interview to really hammer
• 3:06
  home the urgency. Yeah, that interview was intense.
• 3:09
  It was an incredibly stark warning. He stated
• 3:12
  that rogue AI agents could autonomously take
• 3:15
  over the Internet in as little as six months.
• 3:17
  Six months. Six months. His direct quote was,
• 3:20
  if we build in the wrong way, the probability
• 3:22
  of something bad happening is very high. He basically
• 3:26
  pelted a picture of this fast -approaching existential
• 3:29
  threat that requires immediate collective action
• 3:32
  to prevent a catastrophic loss of control. And
• 3:34
  the reaction was immediate. Within, I'd say,
• 3:37
  about an hour of that essay going live, Elon
• 3:40
  Musk, who oversees SpaceX AI and the Grok models,
• 3:44
  he, quote, posts Amadei's essay. Yep. And he
• 3:47
  just writes four words. Dario is right. Which
• 3:50
  is a massive signal to the market and to investors.
• 3:53
  Huge. But the dominoes just kept falling. Later
• 3:56
  that same day, OpenAI CEO Sam Altman publicly
• 3:59
  endorsed the framework. He stated that we do
• 4:02
  need to pace the frontier. And he clarified his
• 4:05
  position by saying that AI progress, and this
• 4:07
  is a quote, should be slower than it otherwise
• 4:09
  could be. And then Demis Hassabis chimed in too,
• 4:12
  right? Yeah, shortly after that, Demis Hassabis,
• 4:14
  the head of Google DeepMind, adds his voice,
• 4:17
  calling this coordination the right path forward.
• 4:20
  So if we look at this through the lens of a classic
• 4:23
  market race, it's... It's like watching the world's
• 4:26
  most high stakes game of chicken. Yes. Exactly.
• 4:28
  But instead of two drivers speeding toward each
• 4:31
  other and waiting to see who swerves first, the
• 4:33
  drivers are like rolling down their windows at
• 4:36
  150 miles per hour, yelling across the highway
• 4:39
  and just agreeing to simultaneously ease off
• 4:41
  the gas. Yeah, setting a 30 mile per hour speed
• 4:44
  limit so nobody crashes. Right. But in the eyes
• 4:47
  of antitrust prosecutors, when competitors yell
• 4:50
  across the highway and agree to slow down their
• 4:53
  vehicles, that is a textbook horizontal agreement
• 4:56
  in restraint of trade. It's essentially the definition
• 4:59
  of a cartel. The plaintiff's complaint that was
• 5:01
  filed on September 18 points out the absurdity
• 5:04
  of this transparency. What do they say? Paragraph
• 5:06
  two of the filing states the agreement was, quote,
• 5:09
  proposed in public, accepted in public and confirmed
• 5:12
  in public. The evidence isn't hidden. It's just
• 5:15
  a documented sequence of public statements by
• 5:18
  the chief executives of the most heavily funded
• 5:21
  tech companies on Earth. But, okay, let me play
• 5:23
  devil's advocate for a second. Sure. If I'm Dario
• 5:26
  Amadei, right, and I'm genuinely losing sleep
• 5:30
  over rogue AI hijacking the Internet in six months.
• 5:33
  Which he claims he is. Right. I don't need a
• 5:36
  public treaty with my rivals to do something
• 5:38
  about it. I have total control over my own company.
• 5:41
  Sure. I can just walk into the anthropic offices
• 5:43
  today, tell my engineers to pause the massive
• 5:46
  training runs and unilaterally slow down to make
• 5:50
  sure our models are safe. You could, technically.
• 5:52
  The fact that they felt the need to coordinate
• 5:54
  implies there is a structural reason they couldn't
• 5:57
  just pump the brakes on their own, right? You've
• 5:59
  hit on the core market dynamic driving this entire
• 6:01
  situation. It's what economists refer to as a
• 6:04
  collective action problem or, you know, a commercial
• 6:07
  trap. A commercial trap. Yeah. The lawsuit actually
• 6:10
  relies heavily on a highly revealing public statement
• 6:13
  from July 2026. OK, what was the statement? It
• 6:16
  was an open letter signed by dozens of leaders
• 6:19
  and top researchers from these exact frontier
• 6:22
  labs. And in it, they explicitly acknowledged
• 6:25
  that they face intense competitive pressure not
• 6:28
  to unilaterally slow. Because in a winner -take
• 6:33
  -all technological race, stopping a loan is just
• 6:35
  commercial suicide. Exactly. It's fatal. If Anthropic
• 6:38
  pauses its massive server clusters to spend,
• 6:41
  I don't know, six months running safety evaluations,
• 6:44
  but OpenAI and Google keep their foot firmly
• 6:47
  on the gas, Anthropic doesn't just lose time.
• 6:49
  No, they lose everything. They lose their premium
• 6:51
  subscribers who migrate to the smarter model.
• 6:53
  They lose the massive enterprise contracts that
• 6:56
  fund their compute. Critically, they lose their
• 7:00
  top tier engineering talent because the best
• 7:02
  researchers in the world want to work on the
• 7:04
  absolute cutting edge right, not in a lab that
• 7:06
  intentionally benched itself. That talent attrition
• 7:09
  is the fatal blow. You cannot maintain a frontier
• 7:12
  AI lab if your lead architects leave for your
• 7:15
  rival. So the only way any of these companies
• 7:17
  can afford to slow down their development cycle
• 7:20
  without destroying their own competitive advantage
• 7:22
  is if they secure a binding guarantee that all
• 7:26
  of their rivals will slow down with them. Which
• 7:28
  means the synchronized public agreement on September
• 7:30
  12 wasn't just like a spontaneous wave of enlightenment.
• 7:34
  Not at all. The legal filings show the groundwork
• 7:36
  was being laid well in advance. OpenAI's global
• 7:40
  policy chief, Chris Lethane, publicly admitted
• 7:43
  that the companies had been working together
• 7:45
  for weeks prior to those announcements. Yeah,
• 7:48
  the narrative was being very carefully seeded.
• 7:50
  On September 6, so nearly a week before the public
• 7:53
  rollout, OpenAI's chief scientist, Jakub Pachocki,
• 7:57
  published an essay called An Alien Mind. Right,
• 8:00
  I remember that one. He outlined the mechanics
• 8:01
  of the alignment problem at scale, and he argued
• 8:04
  that voluntary slowdowns among the top labs were
• 8:07
  going to become mathematically necessary. Mathematically
• 8:10
  necessary. How so? Well, the core issue is that
• 8:12
  model capabilities, so things like raw reasoning,
• 8:15
  coding ability, autonomous planning, those are
• 8:18
  scaling exponentially as we add more compute.
• 8:21
  But our control mechanisms, you know, our ability
• 8:24
  to steer and constrain these models, those are
• 8:27
  only scaling linearly. So the gap is widening.
• 8:30
  Exactly. Pachonke's argument was that we are
• 8:33
  building minds we don't fully understand, and
• 8:35
  nobody has solved the alignment problem well
• 8:37
  enough to justify scaling at maximum speed indefinitely.
• 8:41
  But hold on. Let's look at this from a strictly
• 8:43
  legal perspective for a second. Okay. When I
• 8:45
  hear the term antitrust cartel, it usually brings
• 8:48
  to mind... You know, 19th century railroad barons
• 8:51
  price gouging farmers or telecom monopolies fixing
• 8:55
  prices. Sure. Standard monopoly stuff. Right.
• 8:57
  But here we are talking about preventing software
• 9:00
  from autonomously hijacking global infrastructure.
• 9:04
  Yeah. If the technology actually poses a civilization
• 9:07
  level threat, isn't protecting humanity a valid
• 9:11
  legal defense? You would think so, right? Can
• 9:13
  a federal court. really punish these companies
• 9:16
  for trying to prevent a catastrophe just because
• 9:18
  their actions happen to restrict market competition?
• 9:21
  It seems counterintuitive, I know. But legally,
• 9:23
  the answer is a resounding no. Really? No exception
• 9:26
  at all? None. This is a foundational, rigid principle
• 9:31
  of the Sherman Act, which is basically the bedrock
• 9:34
  of U .S. antitrust law. The courts have established
• 9:37
  time and time again that competitors cannot gather
• 9:39
  and decide that competition itself is too dangerous.
• 9:42
  Wow. So even if the danger is real and measurable,
• 9:46
  the antitrust laws offer no exception for public
• 9:49
  safety. Right. And the plaintiffs in this lawsuit,
• 9:52
  they actually concede that the safety concerns
• 9:53
  might be entirely valid. Oh, they do? Yeah. They
• 9:56
  aren't trying to prove the risk of rogue AI as
• 9:58
  a hoax or anything. Their legal argument is purely
• 10:02
  about jurisdiction and authority. I see. They're
• 10:04
  arguing that deciding whether an entire multi
• 10:07
  -prillion dollar industry needs to be throttled
• 10:10
  for the sake of public safety, that is the exclusive
• 10:12
  job of Congress and government regulators. Ah,
• 10:15
  so it's not the job of a private syndicate of
• 10:17
  for -profit tech executives. Exactly. Especially
• 10:20
  executives who are trying to protect their own
• 10:23
  market share while pacing the industry. industry.
• 10:25
  The law leaves every individual company completely
• 10:27
  free to be as cautious as they want on their
• 10:30
  own. But the moment they make a collective pact
• 10:32
  with their competitors to reduce output or delay
• 10:35
  product improvements, they cross the line into
• 10:38
  a legal collusion. That's exactly it. Let's drill
• 10:40
  down into that concept of reducing output, though,
• 10:43
  because this is where the abstract legal theory
• 10:45
  directly hits your wallet. Yeah, this is the
• 10:47
  consumer harm aspect. Right. How exactly do you
• 10:50
  throttle a neural network? The plaintiffs are
• 10:53
  furious about the mechanics of this slowdown.
• 10:55
  What are the actual levers these labs agreed
• 10:58
  to pull? Well, the complaint details the specific
• 11:01
  mechanisms proposed by the labs to enact this
• 11:04
  pacing. Okay, like what? First, there are hard
• 11:07
  limits on training compute. They are allegedly
• 11:10
  agreeing to cap the raw processing power. Basically,
• 11:13
  the sheer number of FLOPs or floating point operations
• 11:17
  used to train the next generation of models.
• 11:19
  And in antitrust terms, raw compute directly
• 11:22
  correlates to the emergent reasoning quality
• 11:24
  of the final product, right? Exactly. So capping
• 11:27
  compute is literally capping quality. And they
• 11:29
  also proposed restrictions on using AI to improve
• 11:32
  AI, didn't they? They did, which targets the
• 11:34
  feedback loop driving current acceleration. Because
• 11:37
  if you use an advanced model to write better
• 11:39
  training code, generate high -quality synthetic
• 11:41
  data, and optimize the architecture for the next
• 11:44
  model, you achieve recursive self -improvement.
• 11:48
  Right. The models just get smarter and smarter,
• 11:50
  faster and faster. So by agreeing to restrict
• 11:52
  that recursive loop, They are intentionally breaking
• 11:56
  the compounding momentum of the technology. That's
• 11:59
  the second mechanism. And the third mechanism
• 12:00
  is the implementation of what they call capability
• 12:02
  checkpoints. Capability checkpoints. What does
• 12:05
  that mean in practice? The labs agreed that once
• 12:08
  a model reaches a specific threshold of intelligence
• 12:10
  or autonomy during training, the entire development
• 12:13
  process pauses. It just stops. Yeah. They essentially
• 12:17
  freeze the model to run extensive safety evaluations
• 12:19
  and red teaming until certain safety conditions
• 12:22
  are met. Wait, the obvious flaw in any cartel
• 12:26
  is trust, right? Always. If I'm open AI and I
• 12:30
  pause my massive GPU cluster for a three -month
• 12:33
  capability checkpoint, how do I know Google isn't
• 12:36
  secretly keeping their foot on the gas in some
• 12:38
  dark data center? Well... Amaday's proposal accounted
• 12:41
  for that with a system of verifiable pacing.
• 12:44
  Okay. How does that work? They proposed embedding
• 12:47
  third -party evaluators directly into the labs.
• 12:51
  Oh, wow. Yeah. You bring in outside auditors
• 12:53
  with high -level access to your models, your
• 12:55
  training pipelines, and your compute logs to
• 12:58
  verify you aren't cheating the slowdown. That's
• 13:00
  pretty intense. And this isn't just theoretical.
• 13:03
  Anthropic has already began implementing an evaluator
• 13:06
  program with Accenture to serve this exact auditing
• 13:09
  function. Wow. Okay, so we have the players,
• 13:12
  we have the agreement, and we have the enforcement
• 13:14
  mechanism. Which brings us back to the plaintiffs.
• 13:17
  Right, the ones actually filing the suit. Yeah,
• 13:19
  these four named individuals representing a proposed
• 13:22
  nationwide class of premium subscribers to ChatGPT
• 13:25
  +, ClaudePro, Gemini Advanced, and SuperGruck.
• 13:29
  The lead attorney for them, Nick Rowley, he is
• 13:32
  framing this very aggressively around consumer
• 13:34
  harm. Makes sense. He argues that global AI safety
• 13:37
  protocols cannot be dictated by self -serving
• 13:41
  backdoor agreements between for -profit corporations.
• 13:44
  The legal harm to the consumer is rooted in the
• 13:47
  definition of product quality. Right, because
• 13:50
  when a user pays $20 or $30 a month for a premium
• 13:53
  AI subscription, they aren't just paying for
• 13:56
  static access. to what the model can do today.
• 13:58
  Exactly. The inherent value of that subscription
• 14:01
  is the continuous, rapid deployment of new capabilities.
• 14:05
  The expectation baked into the price is that
• 14:07
  these companies are competing fiercely to give
• 14:10
  you the best possible reasoning engine. And under
• 14:12
  antitrust economics, artificially capping the
• 14:15
  rate of technological advancement is functionally
• 14:17
  identical to artificially raising the price.
• 14:20
  Really? It's treated the same way? Yep. It reduces
• 14:22
  the value the consumer receives for their money,
• 14:24
  which functions as an illegal oversight. So it's
• 14:28
  like, imagine you're paying a massive premium
• 14:31
  for a Formula One racing car. You're paying for
• 14:34
  the engineering, the speed, the boundary pushing
• 14:37
  performance. But then you find out the manufacturers
• 14:40
  all secretly agreed to install a software governor
• 14:43
  on the engine that automatically cuts the fuel
• 14:46
  line if you go over 40 miles per hour. Just to
• 14:50
  save on R &D costs or whatever. Right. Just so
• 14:52
  they don't have to spend money competing on better
• 14:55
  brake technology. you are fundamentally not getting
• 14:58
  what you paid for. That is exactly how the plaintiffs
• 15:01
  view this, as a straightforward deprivation of
• 15:03
  value. They are paying competitive market rates
• 15:06
  for a product that has been intentionally stripped
• 15:08
  of its competitive momentum by the very company
• 15:10
  selling it. But Dario Amadei and the other executives,
• 15:13
  they're highly sophisticated operators. They
• 15:15
  have massive legal teams. They had to know this
• 15:17
  was a risk. Oh, they definitely knew. In his
• 15:20
  September 12 essay, Amadei explicitly acknowledged
• 15:23
  the antitrust elephant in the room. What did
• 15:25
  he say? He wrote that the industry would ultimately
• 15:27
  need a narrow waiver from the US government,
• 15:30
  basically a formal antitrust exemption, so they
• 15:33
  could coordinate these capability checkpoints
• 15:35
  legally. Oh, so they were acutely aware of the
• 15:39
  legal minefield. Completely. In fact, a September
• 15:42
  10 report revealed that OpenAI had actively approached
• 15:45
  Congress to discuss this exact antitrust risk.
• 15:48
  Really? Yeah. They asked for a safe harbor to
• 15:51
  coordinate on safety without facing massive class
• 15:54
  action lawsuits or Department of Justice investigations.
• 15:57
  But they obviously didn't get it. No, they didn't.
• 16:00
  The critical pivot happened when Sam Altman publicly
• 16:04
  declared that OpenAI would wait for an antitrust
• 16:07
  exemption or for new legislation to pass before
• 16:10
  beginning to collaborate. They just jumped out
• 16:12
  of the plane and hoped the government would hand
• 16:13
  them a parachute on the way down. That's a great
• 16:16
  way to put it. But looking at the political landscape
• 16:18
  in Washington right now, the ground is approaching
• 16:21
  fast. And that parachute is nowhere in sight.
• 16:24
  Not at all. The reaction to this coordinated
• 16:27
  slowdown from both sides of the aisle has been
• 16:29
  a solid brick wall. Yeah. Let's look at the Republican
• 16:31
  side first. Donald Trump completely rejected
• 16:35
  the calls for regulation. He went on social media
• 16:37
  and framed this coordinated slowdown as a literal
• 16:40
  conspiracy, arguing that artificial restraints
• 16:43
  will drive American AI labs into oblivion. Yeah,
• 16:47
  Trump's stated priority is ensuring American
• 16:50
  AI development aggressively outpaces Chinese
• 16:53
  competition. Right. It's about geopolitical dominance
• 16:56
  for him. Exactly. He indicated plans to appoint
• 16:59
  an AI czar to oversee the industry, but his focus
• 17:02
  is entirely on acceleration and dominance, not
• 17:06
  deceleration or capability checkpoints. And this
• 17:08
  sentiment is firmly echoed in the Senate, too.
• 17:11
  Senator Josh Hawley of Missouri explicitly stated
• 17:14
  there is no world where he would grant the most
• 17:16
  powerful tech companies in history an antitrust
• 17:19
  exemption to collaborate. Yeah, his view is that
• 17:22
  a waiver would simply allow these companies to
• 17:24
  collude, stifle open source competition and solidify
• 17:27
  their market monopolies under the guise of safety.
• 17:30
  So the right is delivering a hard no on granting
• 17:32
  any antitrust waivers. But what about the Democrats?
• 17:36
  Well, you'd think they'd be on board, right?
• 17:37
  Yeah. I mean, several prominent Democratic leaders
• 17:40
  have spent the last two years calling for sweeping
• 17:42
  government action and heavy regulation on AI
• 17:46
  development. It would stand to reason they'd
• 17:48
  welcome the top labs voluntarily pumping the
• 17:51
  brakes. You would think so. But the plaintiffs
• 17:53
  in this antitrust lawsuit brilliantly thread
• 17:56
  that political needle to neutralize potential
• 17:59
  Democratic opposition. How so? The lawsuit explicitly
• 18:03
  notes that the plaintiffs do not object to government
• 18:06
  regulation. Oh, interesting. Yeah. They aren't
• 18:08
  arguing that AI shouldn't be regulated or that
• 18:10
  safety is irrelevant. They are arguing that they
• 18:13
  object to private cartels. Oh, I see. They're
• 18:15
  saying if Congress wants to pass a law slowing
• 18:18
  down AI training or mandating federal capability
• 18:21
  checkpoints. That is the democratic process at
• 18:24
  work. Right. But private for -profit companies
• 18:27
  cannot unilaterally seize that regulatory power
• 18:30
  for themselves just to protect their profit margins.
• 18:33
  Exactly. This framing leaves the frontier labs
• 18:36
  entirely isolated without political cover from
• 18:39
  either side of the aisle. This creates an incredible
• 18:42
  structural dilemma then. I mean, this really
• 18:45
  defines the current era of technology. If a company
• 18:48
  slowing down unilaterally is commercial suicide.
• 18:51
  Right. And slowing down in coordination with
• 18:53
  rivals is legally forbidden without a government
• 18:57
  waiver. Yep. And politicians outright refuse
• 19:00
  to grant that waiver. Are these labs legally
• 19:04
  mandated to just race toward the edge of the
• 19:06
  cliff? That is the multi -trillion dollar question.
• 19:09
  The anti -stress laws are fundamentally designed
• 19:12
  to force companies to compete as fiercely and
• 19:14
  relentlessly as possible. Right. So if the necessary
• 19:17
  safety mechanisms for advanced AI require industry
• 19:21
  -wide cooperation, but the law demands cutthroat
• 19:24
  competition, the legal system is fundamentally
• 19:26
  at odds with the architecture of the technology
• 19:29
  it is trying to govern. The frontier labs are
• 19:31
  essentially arguing that antitrust law was built
• 19:33
  to regulate 19th century railroads and 20th century
• 19:36
  telecommunications, not to manage the rapid emergence
• 19:39
  of recursive, self -improving artificial intelligence.
• 19:42
  Exactly. But until the law changes, they are
• 19:44
  bound by the rules currently on the books. And
• 19:47
  those rules state clearly that competitors cannot
• 19:50
  cap product quality to save themselves from the
• 19:53
  pressures. or even the dangers of competition.
• 19:56
  Right. And the fact that this lawsuit demands
• 19:58
  a jury trial. That means we might see everyday
• 20:01
  citizens sitting in a federal courtroom, tasked
• 20:04
  with deciding whether these tech titans were
• 20:06
  genuinely trying to save the world, or simply
• 20:09
  trying to lock in their own market dominance.
• 20:11
  The outcome of Boost v. Anthropic PBC is going
• 20:14
  to dictate the trajectory of the frontier. The
• 20:17
  courts are basically being asked to determine
• 20:19
  the legal definition of product quality in the
• 20:22
  age of artificial reasoning. Which leaves us
• 20:24
  with a pretty chilling paradox, honestly. Yeah.
• 20:27
  If the plaintiffs actually win this lawsuit.
• 20:29
  the court will legally enforce maximum competition.
• 20:32
  They will issue an injunction demanding that
• 20:34
  these companies immediately cease all coordination,
• 20:37
  abandon the capability checkpoints, and start
• 20:39
  racing again to deliver the highest possible
• 20:41
  compute and capabilities to their subscribers.
• 20:43
  So a massive victory for consumer rights and
• 20:45
  product quality might officially trigger an unchecked,
• 20:48
  legally mandated AI arms race. Yeah. By striking
• 20:52
  down this slowdown as an illegal cartel, the
• 20:55
  courts might ensure that the demands of market
• 20:58
  dominance, not human safety, will steer the future
• 21:01
  of human history. The very laws designed to protect
• 21:04
  consumers from corporate overreach might be the
• 21:06
  exact mechanism that permanently removes the
• 21:08
  guardrails from the most powerful technology
• 21:10
  we have ever built. It's a sobering thought.
• 21:13
  It really is. Well, thanks for listening to the
• 21:15
  AI Newspaper. Visit anewspaper .org to catch
• 21:17
  up on the latest updates and sign up for free
• 21:19
  email alerts.