The AI Brief

Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA), both members of the House Energy and Commerce Committee, released the draft on June 4. The text is built around four pillars: frontier model governance, workforce impact research, cybersecurity posture, and AI research and development funding. The frontier governance provisions require "large frontier developers" (defined as companies with more than $500 million in annual revenue that have trained the most powerful models) to hire licensed independent auditors every six months, publicly disclose risk assessments before each model release, and face civil penalties of up to $1 million per violation per day for non-compliance.

The bill would formally codify the Center for AI Standards and Innovation (the rebranded AI Safety Institute) with $300 million in authorization over three years and require the Government Accountability Office to audit federal AI adoption progress. It defines catastrophic risk precisely: a foreseeable risk of death or serious injury to more than 50 people, or more than $1 billion in property damage, from a model assisting in weapon development, cyberattack, or autonomous harmful action without meaningful human oversight. The three-year preemption of state development laws is the sharpest flashpoint: supporters argue it prevents a compliance patchwork that would slow innovation; critics, including Public Citizen and Americans for Responsible Innovation, argue it dismantles the only active accountability layer currently functioning while federal alternatives remain unbuilt.

The bill arrives days after Trump signed a June 2 executive order establishing voluntary federal agency reviews of new frontier models. The draft text itself explicitly bars any mandatory licensing or preclearance requirement for model releases, a provision that tech industry groups welcomed. Feedback can be submitted to GAAIA@mail.house.gov. No formal introduction date has been set.

The figure comes from Sensor Tower market intelligence estimates reported by Reuters on June 2, not from an OpenAI audited disclosure; it counts monthly active app users, not total users across web and API. The order of magnitude is broadly corroborated. Google Maps, TikTok, Instagram, and YouTube each took five to eight years to reach comparable scale. By the same Sensor Tower reckoning, Anthropic's Claude app had approximately 56 million global monthly active users, a fraction of ChatGPT's base, but growing at roughly 640% year on year against ChatGPT's 62%. Sensor Tower data also found that US ChatGPT users who installed Claude in Q1 2026 spent 5% less time on ChatGPT one month after installation, a modest but measurable substitution signal. OpenAI has separately reported 900 million weekly active users as of February 2026 and $25 billion in annualized revenue. The company is simultaneously redesigning ChatGPT into a "superapp" integrating agents, Codex, and enterprise workflows, with enterprise generating an estimated 40% of revenue and rising.

In a blog post titled "When AI builds itself," published June 4 by the Anthropic Institute, co-founder Jack Clark and Marina Favaro (head of the Anthropic Institute) argued that frontier AI is approaching "full recursive self-improvement" and that the world should have the option to slow or temporarily pause development. The company disclosed that as of May 2026, Claude authored more than 80% of the code merged into Anthropic's own codebase, and that the typical engineer now merges roughly eight times as much code per day as in 2024. A separate internal survey of 130 employees found the median respondent estimated producing roughly four times as much output with AI assistance as without it. Claude shipped over 800 fixes in April 2026 that cut one class of API errors by a factor of roughly a thousand.

Critically, Anthropic did not call for an immediate unilateral halt. The post states a pause would require "multiple well-resourced labs at or near the frontier, in multiple countries, agreeing to stop under the same conditions," with verification that others had actually stopped. Clark told BBC Newsnight that reaching 100% AI-written code is possible within two years. OpenAI responded with a published position that "democratic governments, not private companies acting alone, must ultimately determine the rules." David Sacks, now co-chairing the President's Council of Advisors on Science and Technology, has previously described Anthropic's policy positions as a "regulatory capture agenda," a framing that shapes how the call will be received in Washington. The post arrives the week after Anthropic filed a confidential IPO prospectus with the SEC at a near-trillion-dollar valuation, a timing critics have noted.

Disclosure: Anthropic, mentioned in this item, is the company that develops Claude, which generates this brief.

On June 5 and 6, President Trump publicly confirmed that senior administration officials have been in preliminary discussions with major AI companies about potential government equity stakes, telling Air Force One reporters "there are concepts where pieces could be given to the American public, where the American public essentially becomes a partner." CNBC separately confirmed that OpenAI CEO Sam Altman first pitched the concept to the Trump administration in early 2025 and has revisited it in recent weeks. The proposed structure involves OpenAI donating equity, rather than the government purchasing it, to seed a "Public Wealth Fund" modeled on the company's April 2026 policy proposal. No official investment terms have been decided. Notably, a person familiar with the matter told NOTUS that Anthropic is not participating in these discussions, following a February 2026 Trump executive order placing Anthropic on a national security restricted list after the company declined to support certain Pentagon use cases.

Senator Bernie Sanders has taken a harder line: his forthcoming AI Sovereign Wealth Fund Act would impose a one-time 50% tax on OpenAI, Anthropic, xAI, and similar firms, payable in stock rather than cash, with proceeds placed in a federal sovereign wealth fund granting the government board representation at each company. The bill has not yet been formally introduced. Both proposals face substantial legal and structural obstacles, including constitutional questions about stock seizure and the absence of a federal mechanism for private equity transfers. Critics from the left and the free-market right have found common ground in warning that a government shareholder is also a weakened regulator.

OpenAI released Lockdown Mode on June 5, rolling it out initially to enterprise plans and expanding it to personal accounts and self-serve ChatGPT Business users by June 6. When enabled, the feature disables live web browsing (restricting the model to cached content), agent mode, deep research, image retrieval from the web, Canvas networking, and file downloads. Image generation and manual file uploads remain available. OpenAI explicitly acknowledges that Lockdown Mode does not stop injections from appearing in content ChatGPT processes: a malicious payload embedded in a cached webpage or uploaded PDF can still influence model behavior and response accuracy. The feature addresses only the exfiltration stage, eliminating the outbound network channels an attacker would need to receive stolen data.

Alongside Lockdown Mode, OpenAI introduced "Elevated Risk" labels on specific features that create significant data exposure, such as authorizing an AI agent to access email or a proprietary codebase. These are informational flags, not blocks. Security researcher Simon Willison's framing of the "Lethal Trifecta" (private data access plus untrusted content plus an outbound channel) illustrates why Lockdown Mode is a meaningful but partial control: it removes the third leg. Researchers at The Next Web noted that security teams have documented prompt injection hijacks against agents from Anthropic, Google, and Microsoft via GitHub Actions integrations; all three paid bug bounties but published no public advisories. OpenAI characterizes prompt injection as a "frontier" problem affecting all large language models, and states explicitly that Lockdown Mode is "not intended for everyone."