Anthropic briefs Homeland Security on Mythos AI model amid mounting cybersecurity concerns

Anthropic to brief House Homeland Security panel about Mythos in closed-door meeting

The House Committee on Homeland Security will host Anthropic for a closed-door briefing on its Mythos cybersecurity model on Wednesday, as fears grow in Washington about the safety risks of artificial intelligence. A source familiar with Wednesday's meeting confirmed it to The Hill, stating that the briefing will focus on the "capabilities, national security implications and policy considerations" around Mythos. This marks at least the second briefing in as many weeks that Homeland Security lawmakers have been planning with Anthropic after the limited release of Mythos sent shockwaves through Washington. Mythos, which Anthropic says is its most advanced model to date, can spot decades-old security vulnerabilities, the AI firm claims. The AI model, which Anthropic did not release publicly, can help institutions spot and patch security vulnerabilities more quickly. But it may also be a double-edged sword, empowering hackers to find and potentially exploit these flaws. The Hill has reached out to Anthropic for comment. It comes about two weeks after Anthropic and OpenAI hosted closed-door briefings before the committee on their cybersecurity models. This week's briefing comes as President Trump and a host of other technology and business leaders head to China for bilateral talks with Chinese President Xi Jinping. AI, along with its risks, is expected to be discussed as the White House scrambles to rethink its AI policy in the wake of Mythos.

Anthropic To Brief Homeland Security On 'Mythos' AI Cyber Model

Anthropic will meet behind closed doors with the House Committee on Homeland Security today to walk lawmakers through Mythos, the company's cybersecurity-focused AI system. The session lands as Washington's anxiety about artificial intelligence safety and national security keeps climbing. Three people familiar with the plans reported that the discussion is expected to center on Mythos' "capabilities, national security implications, and policy considerations." The briefing is scheduled as a private meeting with the panel. Anthropic representatives expected to lead the presentation include Logan Graham, head of Anthropic's Frontier Red team, and Josh Tilstra, who works in the company's national security programs and policy. Mythos has not been released to the public, but Anthropic has described it as its most advanced model and says it can surface long-standing software weaknesses. The Pentagon's tech chief, Emil Michael, highlighted the potential security risks of Anthropic's Claude models. He also noted the advanced cyber capabilities of Mythos, another AI model, which could potentially identify and rectify cyber vulnerabilities. Officials have argued that earlier access could help surface risks that span cyber intrusions and possible military misuse. The deleted page had described the testing as a way to spot those problems before models reach the public. The meeting is at least the second such engagement between the committee and Anthropic following a limited rollout of Mythos. Last month, Anthropic and OpenAI also held private briefings for the committee focused on AI models and their implications for cybersecurity. Photo: Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.