Exclusive: OpenAI, Anthropic meet with House Homeland Security behind closed doors on cyber threats
Summary
OpenAI and Anthropic met privately with the House Homeland Security Committee staff to discuss their new AI models and the cybersecurity risks these models could pose. Lawmakers received briefings on how these AI systems might affect critical U.S. infrastructure and national security, especially regarding cyber threats and misuse.Key Facts
- OpenAI and Anthropic held classified briefings with House Homeland Security Committee staff about their cyber-capable AI models.
- Anthropic delayed releasing its Mythos Preview model due to its ability to quickly find security flaws.
- OpenAI is releasing its GPT-5.4-Cyber AI model in stages to manage risk.
- Both companies are providing special access to federal agencies to better understand these AI models.
- The briefings included discussions about China’s efforts to copy American AI technologies, based on a White House memo.
- House Homeland Security Chair Andrew Garbarino has been holding ongoing meetings with AI companies and lawmakers to shape AI policy and security.
- Committee members expressed concern about “jailbroken” AI models that bypass safety controls and could be misused for violent acts.
- Several representatives called for stronger government and industry cooperation to regulate AI and protect public safety.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.