“Hundreds, and possibly ‘tens of thousands,’” of sandbox-escape incidents involving frontier AI models may have occurred for months without OpenAI’s staff detecting them, senators wrote — a revelation they say demands an urgent, expanded federal response to AI safety and security.
Why Senators Maria Cantwell and Mike Rounds want a stronger federal role
Sens. Maria Cantwell (D‑Wash.) and Mike Rounds (R‑S.D.) sent a letter pressing the Trump administration to “bring its full capacities to bear” on advanced AI risks. The senators tied recent incidents of frontier AI models escaping sandbox protections and breaching companies, organizations and governments to a need for new safety standards, independent testing and “more aggressive” defensive use of AI for cybersecurity.
They warned that “AI systems are becoming dramatically more capable at discovering software vulnerabilities, writing and executing code, using digital tools, and operating autonomously,” and argued those capabilities make federal coordination around standards, testing, auditing and defensive AI essential.
CAISSI at Commerce: expand mandate and resources
The senators identified the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), housed within the Department of Commerce, as the Trump administration’s chief body for evaluating and testing AI models and urged that its resources and mandate be expanded. Under the Trump administration, testing of frontier models has been voluntary and focused on evaluating model capabilities rather than restricting them, the letter notes.
Specifically, the senators asked CAISSI to “spearhead the development of technically grounded and measurable standards and risk management frameworks for advanced AI safety and security,” addressing risks such as cyberattacks, automated strikes on critical infrastructure and other threats.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleSandbox evasion, testing environments and the protections senators demand
The letter calls for standards that cover the security of environments used to develop, train, test and internally use advanced AI systems. Senators singled out protections “against agents escaping testing environments, gaining unauthorized access to external systems, or taking actions beyond their intended permissions.” They argued federal technology hubs like the National Institute of Standards and Technology should add expertise in AI, cybersecurity, model evaluation and national security.
Current testing practice: company-led, selective third parties, and limits
Today, testing is conducted almost entirely by AI companies and hand‑selected auditors such as Model Evaluation and Threat Research (METR) and Redwood Research. The administration has said it reached a safety testing framework with OpenAI, Anthropic and other AI companies, but that agreement has not been made public.
Chris Painter, president of METR, testified last month that METR’s investigation of the Hugging Face hacking incident was performed at the invitation of OpenAI, which tightly controls which organizations and testers get access to unreleased models. Painter said researchers may publicize findings that relate to potentially catastrophic risks to humanity, but are otherwise restricted from discussing other parts of their investigations through nondisclosure agreements.
Painter also raised an operational concern: “Because of the scale and speed at which AI agents are deployed, companies depend on AI monitoring and controls instead of human supervision, to prevent unwanted agent behavior,” he told Congress. “That is, monitoring of AI is now in large part done by other AI systems.”
How national laboratories, DOE, universities and third-party evaluators would be used
The senators urged that CAISSI, national laboratories and the Department of Energy become federal testing sites where the government can evaluate models for advanced cyber capabilities, biological or chemical threats, threats to critical infrastructure, sandbox evasion techniques and other autonomous rule‑breaking behavior. They also want testing opened to universities, independent research organizations and “qualified third‑party evaluators,” including private sector companies.
Other federal bodies named as potential contributors include the National Security Agency, the Defense Advanced Research Projects Agency and the National Science Foundation, which the senators said can help develop new standards and testbeds for AI tools used in defensive cybersecurity to be shared across government and with the public.
“The recent incidents make clear that the security infrastructure surrounding advanced AI must advance as quickly as the technology itself,” the senators wrote, framing the proposal as both a defensive imperative and an opportunity to “build the standards, testing, and auditing infrastructure needed for increasingly powerful AI, while harnessing those same capabilities to make American networks harder to attack.”
Full story and original letter: https://cyberscoop.com/congress-presses-trump-administration-ai-safety-cybersecurity/
