Latest / Neural Newscast / OpenAI Faces Florida Suit and Trump Signs AI Order [Model Behavior]
Transcript
- [00:00] Nina Park: From Neural Newscast, this is Model Behavior, [00:03] Nina Park: AI-focused news and analysis on the models shaping our world. [00:12] Nina Park: I'm Nina Park. Welcome to Model Behavior. Today is June 2nd, 2026. This program examines [00:21] Nina Park: how AI systems are built, deployed, and operated in professional environments. We look specifically [00:27] Announcer: at the technical foundations and the regulatory frameworks that govern them. [00:32] Nina Park: I'm Thacher Collins. Today we are looking at a day of major regulatory moves and model releases, [00:38] Nina Park: starting with a significant legal escalation in Florida and a new federal framework [00:43] Nina Park: for oversight that seeks to balance national security with national innovation. [00:47] Announcer: We begin with the news out of Florida. Earlier today, State Attorney General James Uthmeyer filed [00:53] Announcer: a civil lawsuit against OpenAI and its CEO Sam Altman. This follows a criminal investigation [00:59] Announcer: launched just last month. According to reports from Tech Radar, the state claims [01:04] Announcer: ChatGPT was marketed to consumers while OpenAI purportedly deceived the public about the true nature [01:11] Announcer: and dangers of the product. The suit includes very specific allegations connecting [01:15] Announcer: the careless introduction of the chatbot to mental health issues, loss of critical thinking, [01:21] Announcer: and the aiding of criminals. Thatcher, the most notable part of this filing seems to be the attempt [01:26] Announcer: to hold Sam Altman personally liable for these corporate outcomes That is exactly what makes [01:31] Announcer: this case unusual, Nina. The Florida Attorney General is citing Altman's reckless [01:36] Announcer: and willful conduct as a founder. The complaint even references a New Yorker profile from [01:41] Announcer: April 2026 that claims Altman has a disregard for the consequences of deception. [01:47] Announcer: But we have to look at the feasibility here. Can a state realistically hold a CEO [01:53] Announcer: personally accountable for the societal side effects of a generative tool? [01:56] Announcer: OpenAI's existing safety page already outlines how they align models with human values, [02:01] Announcer: and they have been doing this since the explosion of ChatGPT four years ago. [02:06] Announcer: The state's claims about aiding and abetting criminal acts seem like a very high bar to prove in [02:11] Announcer: a civil court. Uthmeyer's argument is that the success of the company was not earned [02:16] Announcer: but rather built on a web of deceit. The suit specifically mentions that minors became addicted [02:22] Announcer: to the tool and that professionals suffered public humiliation through hallucinations. [02:26] Announcer: While OpenAI has not issued a formal response to this specific civil filing yet, [02:30] Announcer: they have historically pointed to their extensive safety evaluations and system-level guardrails. [02:36] Announcer: However, the Florida Attorney General argues these tools did not exist when the product [02:40] Announcer: first scaled, leaving a gap of accountability that they are now trying to fill retrospectively. [02:45] Announcer: They claim the company prioritized growth over the safety of the users. [02:49] Thatcher Collins: And that retrospective angle is where I find some friction in the logic, Nina. [02:54] Thatcher Collins: We are talking about a technology that evolved faster than anyone including its creators expected. [03:00] Thatcher Collins: Blaming OpenAI for a lack of safety standards in 2022 and 2023 when the entire industry [03:05] Thatcher Collins: was still defining what an AI safety standard looked like is a tough legal sell. [03:11] Thatcher Collins: It is like blaming the first car manufacturers for not having three color traffic lights. [03:16] Thatcher Collins: As noted in some of our source material today, those signals did not arrive until years [03:21] Thatcher Collins: after the automobile was already on the streets. It feels [03:24] Thatcher Collins: like this lawsuit is trying to treat a systemic societal shift as a specific corporate crime. [03:30] Announcer: It is a tension we see often, the pace of innovation versus the pace of litigation. [03:36] Announcer: Moving from the courtroom to the development lab, Microsoft had a massive day [03:41] Announcer: at its Build 2026 conference. The Verge is reporting that Microsoft has launched [03:46] Announcer: its first advanced reasoning model developed entirely in-house called MAI Thinking One. [03:54] Announcer: This is a significant pivot. For years, Microsoft relied almost exclusively on OpenAI's models, [04:00] Announcer: but they recently renegotiated their deal to allow for more independence. [04:05] Announcer: This move suggests Microsoft is ready to control its own technological destiny rather [04:10] Announcer: than relying on a partnership that is increasingly complicated by legal and regulatory scrutiny. [04:16] Thatcher Collins: This really changes the dynamic of the Microsoft OpenAI partnership. [04:20] Thatcher Collins: MAI Thinking One is described as a medium-sized model that reportedly matches [04:25] Thatcher Collins: leading flagship models on software engineering benchmarks. Microsoft is stressing [04:30] Thatcher Collins: that they trained this from the ground up on clean data, specifically avoiding distillation [04:34] Thatcher Collins: from third-party models. Does this mean Microsoft is becoming a direct competitor [04:40] Thatcher Collins: to the very company they have invested billions in? It seems they are building [04:44] Thatcher Collins: an internal alternative to GPT-5 or whatever comes next. If they can offer a similar level [04:50] Thatcher Collins: of reasoning at a lower cost, it puts huge pressure on OpenAI to justify its own enterprise pricing. [04:57] Announcer: It certainly looks like they are hedging their bets, Thatcher. Along with MAI Thinking One, [05:03] Announcer: they announced six other models today, covering everything from transcription to image editing. [05:10] Announcer: MAITranscribe 1.5 is reportedly five times faster than its competitors. [05:16] Announcer: They're also integrating a new coding model, MAICodeOne Flash, directly into GitHub Copilot. [05:23] Announcer: This isn't just one model, it is an entire ecosystem of in-house silicon [05:28] Announcer: and software that reduces their reliance on external APIs. This integration allows for [05:33] Announcer: a much tighter feedback loop between the hardware and the software, potentially leading to [05:38] Announcer: the kind of efficiency gains that general purpose models simply cannot achieve. [05:43] Nina Park: But we should be [05:43] Nina Park: careful with the advanced reasoning label. We've seen other reasoning models before, [05:49] Announcer: and they often come with a high latency cost. [05:52] Nina Park: Microsoft [05:53] Announcer: claims this one [05:53] Thatcher Collins: matches leading benchmarks, but we need to see how it performs in real world enterprise environments [05:59] Thatcher Collins: where reliability is more important than a synthetic score. It is also interesting to note [06:04] Thatcher Collins: that they are targeting the medium size market, which suggests they are going for efficiency [06:08] Thatcher Collins: over raw parameter count. In an era where data center costs are spiraling, a smaller, [06:14] Thatcher Collins: smarter model that runs efficiently [06:16] Thatcher Collins: on Microsoft's own Cobalt chips could be the definitive competitive edge in the enterprise market. [06:22] Announcer: The efficiency theme continued with another major announcement, WebIQ. [06:28] Announcer: CNet reports that Microsoft has built a suite of APIs specifically designed for AI agents. [06:34] Announcer: Instead of a search engine for humans like Bing, WebIQ is a search engine for bots. [06:40] Announcer: It delivers search results in a condensed, token-efficient format that allows AI agents to ground [06:45] Announcer: their responses in real-time data without inflating the cost of the query. [06:50] Announcer: This is a fundamental change in how the Internet is indexed and accessed, [06:54] Announcer: prioritizing machine readability over human visual layout and advertisement placement. [06:59] Announcer: This is a critical infrastructure move, Mina. If most online searches [07:03] Announcer: are eventually going to be done by agents rather than people, [07:07] Announcer: which Microsoft predicts will happen soon, then the architecture of the web has to change. [07:12] Announcer: WebIQ is supposedly 2.5 times faster faster than current solutions, with a response time of [07:18] Announcer: under 165 milliseconds. My question is how does this affect the open web? If agents are scraping [07:25] Announcer: and summarizing via these APIs, do the original content creators see any of that traffic? [07:31] Announcer: This seems to push us further toward a headless Internet [07:34] Announcer: where data is just a commodity for models rather than a destination for people. [07:39] SPEAKER_03: That leads us directly into the federal response [07:42] Announcer: to these very same frontier capabilities. This morning President Donald Trump signed [07:47] Announcer: a new executive order regarding AI oversight and national security. According to [07:53] Announcer: the Associated Press, the order establishes a framework for the federal government to vet [07:58] Announcer: the most advanced AI models for up to thirty days before they are released to the public. [Image: [08:02] Announcer: The President of the United States] Thatcher this order was actually delayed from last month [08:06] Announcer: because the President was concerned it might slow down American innovation [08:10] Announcer: or give an advantage to international competitors in the race for digital supremacy. [08:15] Thatcher Collins: Yes it is a very delicate balance. The President was clear that he did not want to hand an advantage [08:21] Thatcher Collins: to China. To address that, the final version of the order is strictly voluntary. [08:26] Thatcher Collins: It asks companies like Google, Anthropic, and OpenAI to collaborate with the government to assess [08:31] Thatcher Collins: advanced cyber capabilities. If a model can find vulnerabilities in critical infrastructure, [08:36] Thatcher Collins: the government wants a 30-day head start to patch those holes before the model goes wide. But Nina, [08:42] Thatcher Collins: if it _is_ voluntary, does it actually have any teeth? Or is it just a way [08:47] Thatcher Collins: for the administration to say they are doing something without actually imposing new regulations? [08:52] Announcer: The administration is framing it as a common sense partnership rather [08:56] Announcer: than a mandatory licensing regime. The order explicitly bars the government from creating [09:01] Announcer: pre-clearance requirements or permits however, it gives the Director of the National Security Agency [09:07] Announcer: significant discretion in deciding which models qualify for scrutiny. As NBC News reports, [09:14] Announcer: this follows the release of Anthropic's Mythos preview model, which apparently showed [09:18] Announcer: a superhuman ability to find flaws in software. The Treasury and the NSA are clearly worried [09:24] Announcer: about what happens when a model like that falls into the wrong hands, Thatcher, [09:28] Announcer: and they want the ability to monitor these capabilities in real time. [09:32] Thatcher Collins: That mythos model really seems to be the catalyst here, Nina. Treasury Secretary Scott Besant [09:39] Thatcher Collins: and Vice President JD Vance have been increasingly vocal about governing the spread of AI to protect [09:44] Thatcher Collins: the economy and national stability. But there is a political friction here too. [09:49] Thatcher Collins: The order directs the Attorney General to prioritize prosecuting crimes involving AI agents. [09:56] Thatcher Collins: It is a shift from regulating the math to regulating the act. I wonder if a 30-day window is [10:02] Thatcher Collins: even enough for the government to secure critical infrastructure against a model [10:06] Thatcher Collins: that can think faster than their human analysts? It feels [10:09] Thatcher Collins: like they are trying to apply a traditional timeline to a non-traditional technological risk. [10:15] Announcer: That is the central concern. Critics, like those cited from the Cato Institute, [10:21] Announcer: worry about giving the NSA too much discretion, calling it a dangerous precedent [10:26] Announcer: that could be weaponized against specific companies. This is especially relevant given [10:31] Announcer: that Anthropic is currently in a legal battle with the administration [10:34] Announcer: over a Pentagon contract dispute. Despite that, OpenAI has already called the executive order [10:40] Announcer: an important step for accountability. They seem eager to position themselves [10:44] Announcer: as the responsible partner in this dynamic. Thatcher, [10:48] Announcer: it appears the industry is split on whether this is helpful oversight or a form of soft censorship. [10:54] Thatcher Collins: It is a complicated dance between Washington and Silicon Valley. While the government tries to wrap [10:59] Thatcher Collins: its head around the risks, OpenAI is pushing ahead with product integration. Today, [11:06] Thatcher Collins: 9to5Mac confirmed that OpenAI is putting its CORDX engine directly into the ChatGPT app [11:12] Thatcher Collins: and releasing six new business-focused plugins. They are trying to move CORDX from a developer tool [11:18] Thatcher Collins: to a general productivity tool for everyone. By making high-level coding logic available [11:23] Thatcher Collins: through a simple interface, they are effectively turning every employee [11:28] Thatcher Collins: into a potential software architect, which could fundamentally change the nature of technical work. [11:34] Announcer: The new plugins are highly specialized: sales, data analytics, creative production, product design, [11:41] Announcer: and even investment banking. The goal is to allow ChatGPT to not just write about a task, [11:47] Announcer: but to use these agentic plugins to actually perform the work in tools like Salesforce, Tableau, [11:53] Announcer: or Figma. They also introduced a sites feature where Codex can turn a text-based plan [11:58] Announcer: into a fully hosted interactive dashboard or app instantly. This move toward action-oriented AI [12:05] Announcer: suggests that the time spent on manual software configuration [12:08] Announcer: and dashboard building may be coming to an end for many professionals. [12:13] Thatcher Collins: The sites feature is particularly interesting because it turns the large language model [12:18] Thatcher Collins: into a UI builder. But again, Nina, we see this recurring theme, the agent taking action. [12:25] Thatcher Collins: OpenAI is clearly trying to ensure that ChatGPT remains the primary interface for business, [12:31] Thatcher Collins: especially as Microsoft builds its own competing MI models. It feels like the era of the chatbot [12:36] Thatcher Collins: is officially over, and we are now in the era of the autonomous workspace. [12:41] Thatcher Collins: The challenge for these companies now is not just generating text but ensuring [12:46] Thatcher Collins: that these autonomous actions are safe, reversible, and fully auditable by the human supervisors. [12:52] Announcer: And the financial markets are reacting accordingly to this massive shift in utility. Yesterday, [12:58] Announcer: June 1st, Anthropic confidentially filed for an IPO. Reports from CNBC [13:04] Announcer: and Medium indicate the company is carrying a valuation of $965 billion. This comes after a massive [13:12] Announcer: $65 billion funding round and a staggering revenue jump from $9 billion at the end of last year to [13:19] Announcer: $47 billion by May 2026. OpenAI is reportedly right behind them with its own confidential S1 filing, [13:27] Announcer: also targeting a valuation north of a trillion dollars. [13:31] Announcer: These numbers represent an almost unprecedented acceleration in enterprise value. [13:36] Thatcher Collins: Those numbers are almost hard to process, Nina. To go from $9 billion to $47 billion in revenue in [13:43] Thatcher Collins: less than six months is unprecedented in the history of enterprise software. [13:49] Thatcher Collins: It shows that over a thousand enterprise clients are now spending more than a million dollars a year [13:55] Thatcher Collins: on Cloude. This IPO filing isn't just a fundraising event. It is a structural shift. [14:01] Thatcher Collins: Anthropic is moving from a high-growth research lab to an institutional infrastructure provider. [14:06] Thatcher Collins: My question is, how does this change their relationship with the companies building on their models? [14:11] Thatcher Collins: Public market scrutiny often changes the priorities of even the most mission-driven companies. [14:16] Announcer: It means the vendor risk models have to change, thatcher. Public companies face [14:21] Announcer: quarterly earnings pressure and margin scrutiny. If you are a chief information officer [14:25] Announcer: and your entire compliance or customer service stack is built on cloud, you now have to worry about [14:31] Announcer: whether a public anthropic will raise token prices to satisfy shareholders. [14:35] Announcer: The grow-at-all-cost phase is ending and the institutional rigor phase is beginning. [14:40] Announcer: This transition will require these companies to prove that their massive valuations are supported by [14:45] Announcer: sustainable long-term business models rather than just the promise of future breakthroughs. [14:50] Thatcher Collins: It also signals that the market [14:52] Nina Park: sees the real value in orchestration. [14:55] Thatcher Collins: We are no longer talking about who has the best chatbot, but who powers the most effective agents. [15:03] Thatcher Collins: As the source material mentions, the race to a trillion dollars isn't just [15:07] Thatcher Collins: about who is the smartest, but who becomes the cognitive engine for the next decade of software. [15:14] Thatcher Collins: With SpaceX also looking to debut next week, 2026 is shaping up to be the year AI infrastructure [15:21] Thatcher Collins: goes public. It is a pivotal moment for the industry [15:25] Thatcher Collins: as these technologies move from the lab to the core of the global financial system. [15:30] Announcer: Thank you for listening to Model Behavior, a Neural Newscast editorial segment. Neural Newscast [15:39] Announcer: is AI-assisted, human-reviewed. View our AI transparency policy at neuralnewscast.com. [15:46] Nina Park: This has been Model Behavior on Neural Newscast. Examining the systems behind the story.