OpenAI, Anthropic Brief House Panel on Cyber AI
SkimNews Take
The closed-door briefings on AI cyber capabilities suggest a proactive effort to integrate private sector technological advancements into national security frameworks before widespread public deployment.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI and Anthropic held two separate classified briefings for House Homeland Security Committee staff on Thursday, covering their new cyber-capable AI models; OpenAI said the session was one of several it held last week with Senate, House, and White House officials.
- The briefings also addressed a recent White House memo accusing China of "industrial-scale" campaigns to distill and copy American AI models.
- Anthropic has withheld public release of its Mythos Preview model because it can quickly find and exploit critical security flaws, while OpenAI used a tiered release approach for its GPT-5.4-Cyber model; both companies are working with federal agencies on access.
- In a separate classified briefing last week on jailbroken AI models, members saw demos of tools being used to plot school shootings and bombings; Rep. August Pfluger (R-TX) called it "very scary" and said the models are "very powerful."
- Rep. Andy Ogles (R-TN) said the jailbreak demo was "rather frightening" because most of the tools were "readily off the shelf and easy to access," and warned "Congress is light years behind."
- House Homeland Security Chair Andrew Garbarino (R-N.Y.) has been hosting ongoing private roundtables with AI executives, and Rep. Jay Obernolte (R-Calif.) introduced a bipartisan federal AI framework bill this week.
Why it matters: Anthropic and OpenAI voluntarily walked Congress through their frontier cyber models and jailbreak exploits, with Rep. Ogles warning 'Congress is light years behind.' The demos of weaponization scenarios appear to be accelerating momentum for Obernolte's newly introduced federal AI framework bill.
