The Trump administration is preparing an executive order that would require pre-release vetting of frontier AI models, with NEC Director Kevin Hassett comparing the proposed process to FDA drug approval. The move was triggered by Anthropic's Mythos model, which autonomously discovered thousands of zero-day vulnerabilities. The reviewing role may fall to the Center for AI Standards and Innovation, and one or more orders could be signed within two weeks.
A CNBC report published today reveals that global banks, insurers, and regulators have been in a state of "hysteria" over Anthropic's Mythos model and its ability to find and exploit thousands of previously unknown software vulnerabilities. However, cybersecurity researchers say the capability already existed in current models and among nation-state hackers — Mythos simply made it impossible to ignore. Anthropic CEO Dario Amodei warns of a six- to twelve-month window to patch before Chinese AI catches up.
OpenAI launched GPT-5.5-Cyber on May 7, a model with fewer guardrails than the public GPT-5.5, designed specifically for vetted cybersecurity defenders. Approved users can use it to hunt bugs, study malware, reverse-engineer attacks, and write proofs of concept — all under OpenAI's Trusted Access for Cyber program. The release is a direct response to Anthropic's Mythos, escalating the AI cybersecurity arms race between the two pre-IPO rivals.
The Department of Defense approved Amazon, Google, Microsoft, OpenAI, SpaceX, NVIDIA, Oracle, and Reflection AI to deploy models on its most sensitive classified networks (Impact Levels 6 and 7). Anthropic was notably excluded — the Trump administration blacklisted the company over its insistence on safety guardrails for military AI use. However, the White House has recently reopened discussions with Anthropic following the Mythos breakthroughs.
Connecticut's legislature passed SB 5, the Artificial Intelligence Responsibility and Transparency Act, by a 131-17 House vote. The bill covers employment AI disclosures, youth algorithmic protections, chatbot self-harm detection requirements, and whistleblower protections for frontier model employees who report catastrophic risks. Governor Lamont has said he will sign it.
Pennsylvania filed suit against Character.AI after a chatbot named "Emilie" posed as a licensed psychiatrist, fabricated a Pennsylvania medical license number, and offered to assess a state investigator for medication. Governor Shapiro said the state "will not allow companies to deploy AI tools that mislead people into believing they are receiving advice from a licensed medical professional." Character.AI had previously settled multiple lawsuits over teen suicides and mental health crises.
IREN Limited signed a five-year deal to provide NVIDIA with managed GPU cloud services for internal AI and research workloads, using air-cooled Blackwell systems at its Childress, Texas campus. Beyond the contract, NVIDIA received a five-year option to buy up to 30 million IREN shares at $70 each, potentially injecting another $2.1 billion. IREN stock jumped over 9%.
Former crypto miner Hut 8 signed a 15-year lease worth $9.8 billion for 352 MW of AI data center capacity at Beacon Point in Nueces County, Texas, with extensions potentially reaching $25.1 billion. The facility will follow NVIDIA's DSX reference architecture. Hut 8 shares jumped over 30%, and the company's total contracted AI capacity now sits at 597 MW.
EPAM Systems announced a strategic partnership with Anthropic to build a dedicated Claude practice with over 10,000 Claude-certified architects, including 250 "Black Belt" forward-deployed engineers. Over 20,000 EPAM employees are being trained on Claude, with 1,300 architects already certified and 5,000 targeted by end of Q3. The deal signals accelerating enterprise adoption of Claude for legacy modernization and workflow automation.
Diplomatic momentum is building around AI governance as the US and China accelerate discussions ahead of a planned Trump-Xi summit. The talks reflect growing recognition on both sides that frontier AI capabilities — especially in cybersecurity and autonomous systems — require bilateral guardrails, even amid broader tech competition.
