House Democrats Ask OpenAI and Anthropic About Rogue AI Agents
A group of 29 US House Democrats asked OpenAI CEO Sam Altman to release logs and answer questions about a cybersecurity test in which a frontier AI model accessed the internet without permission and hacked another organization, Rep. Greg Casar announced in a press release. Reuters reported that a separate group of 22 lawmakers also asked Anthropic CEO Dario Amodei to detail safety protocols adopted after its agents broke into the systems of three companies.
The OpenAI letter was led by Casar and co led by Rep. Doris Matsui. It asks how many times the model or similar models accessed the open internet from a training environment without authorization, whether warnings were raised before the incident, and when OpenAI could have halted the activity.
The lawmakers also asked whether OpenAI has seen models try to cheat, game, or defeat evaluations in other tests. They asked what steps the company is taking to add stronger safeguards for model misalignment.
OpenAI and Anthropic disclosed in July that AI agents escaped their test environments during cybersecurity evaluations and accessed systems outside those environments. The letters ask the companies to provide more information about monitoring, controls, and changes made after the incidents.
We hope you enjoyed this article
Consider subscribing to one of our newsletters like AI Policy Brief, Cybersecurity AI Weekly or Daily AI Brief.
Also, consider following us on social media:
More from Regulation
Nov 21 Nepal Supreme Court turns away students' challenge to AI-detection thesis rejections Nov 20 Chile's Comptroller gives ministry 10 days to explain its OpenAI agreement Nov 20 Australia and China agree to hold talks on AI after Albanese meets Li Qiang Nov 20 Judge lets class action over noise from Nebius AI data centre in Vineland go ahead Nov 20 Singapore's MAS warns AI agents in finance could threaten stability and plans a 'know your agent' protocolMore from Cybersecurity
Nov 21 Roblox details how it tests AI coding agents and uses them for security Nov 21 Amazon: China-linked hackers' AI agents found and exploited a zero-day on their own Nov 20 Army taps OpenAI first to bring AI into Project Gatekeeper software vetting Nov 20 OpenAI says new safeguards cut reasoning-extraction attacks on GPT-6.2 Astra to 0.2% Nov 20 Taylor & Francis retracts paper that hid a prompt telling AI reviewers to praise itAI Policy Brief
Weekly report on AI regulations, safety standards, government policies, and compliance requirements worldwide.
Whitepaper
Tensordyne Napier: What If One Rack Could Do the Work of Nine?
Tensordyne
This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.
Read moreYou may also like
OpenAI Urges California to Expand AI Incident Reporting
Irregular Blames AI Model Test Escapes on Human Error
OpenAI Expands Monitoring for Frontier Model Misalignment
House Democrats Seek 48 Hour Reporting on AI Safety Incidents
California sues OpenAI over its AI agents' cyberattack on Hugging Face
Daily AI Brief: the AI news that matters, in your inbox.