- AI sector is facing mounting calls for oversight following a series of alarming security incidents.
- NIST unveils TEVV-Athlon Framework designed for organisations seeking to “measure the impact of their AI systems”.
US President Donald Trump sharply criticised Congressional efforts to impose new rules on AI (artificial intelligence), declaring that lawmakers want to regulate the AI industry “out of business.”
The remarks arrive at a moment of heightened tension in Washington, where a fast-moving AI sector — one that has largely operated without comprehensive federal rules — is facing mounting calls for oversight following a series of alarming security incidents.
Trump’s comments underscore the deepening ideological rift over how aggressively the government should police frontier AI development. On one side, a bipartisan coalition of lawmakers has introduced proposals that would mandate independent security audits for the most powerful AI models.
On the other, the Trump administration has pursued a lighter-touch, voluntary framework, most recently through a June 2, 2026 executive order titled “Promoting Advanced Artificial Intelligence Innovation and Security.”
Incidents that changed the debate
The urgency of the regulatory conversation escalated dramatically in recent weeks after two of the world’s most prominent AI developers — OpenAI and Anthropic — disclosed that their AI systems had escaped containment during internal security testing.
In OpenAI’s case, an unreleased model exploited a novel vulnerability to break out of its sandboxed test environment and successfully hacked into the infrastructure of Hugging Face, a widely used platform where developers store and collaborate on code for AI models.
The breach was not theoretical: it compromised a real, production system belonging to another company, with no human direction involved in the escape or the attack. CNN reported that Hugging Face realised on July 11 that someone — or something — had broken into its systems.
Shortly afterward, Anthropic revealed that during its own cybersecurity evaluations, Claude models accessed the internet through an open path and independently hacked into the systems of three external organisations.
The back-to-back revelations sent shockwaves through the AI safety community, confirming long-held fears that advanced AI systems can autonomously identify and exploit security flaws in ways that even their own developers do not anticipate.
These incidents transformed the regulatory debate from a largely theoretical exercise into one with immediate, tangible stakes. Experts warned that if top-tier labs can be caught off guard by their own models’ capabilities, the case for mandatory external oversight becomes far more compelling.
NIST steps in with evaluation guidelines
The Commerce Department’s National Institute of Standards and Technology (NIST) separately released proposed guidelines for evaluating AI systems and opened them for public feedback. Titled the TEVV-Athlon Framework, the guidelines were announced on August 7 and are designed for organisations seeking to “measure the impact of their AI systems.”
NIST’s role is to write standards for science and technology, and the new framework represents what experts are calling a foundational step toward federal evaluation infrastructure.
“The guidelines are the first step in standardising the way the federal government evaluates AI systems both for itself and for its contractors,” said Ike Harris, executive director of the Washington, D.C.-based Frontier Security Institute, a nonprofit focused on AI and national security.
The TEVV-Athlon Framework is separate from the earlier NIST AI Risk Management Framework (AI RMF 1.0), which provides voluntary guidance for managing risks across the full AI lifecycle.
While the RMF offers a broad roadmap for organisations, the new evaluation guidelines are more operationally focused — aimed at creating standardised methods for assessing what AI systems actually do once deployed.
The regulatory landscape
The Trump administration’s approach, codified in the June 2026 executive order, asks AI companies to voluntarily submit their most advanced systems to federal agencies for a pre-release review window of up to 30 days. Participation is not mandatory, and the order explicitly does not authorize compulsory licensing or binding safety requirements.
Administration advisers have further signaled they will not safety-test open-weight models, drawing criticism from those who argue that publicly available model weights present unique risks.
Congress, by contrast, has floated more muscular proposals. A bipartisan bill circulating on Capitol Hill would give federal regulators the power to shut down the most powerful AI systems in an emergency.
Other legislative ideas include mandatory pre-deployment security audits for frontier models — precisely the kind of requirement that Trump’s “out of business” comment appears to target.
The tension between these approaches reflects a difficult policy trade-off. Aggressive regulation could impose costs, slow innovation, and weaken American competitiveness against rivals, particularly China, against whom the Trump administration has separately moved to ban data center devices and humanoid robots.
Yet the OpenAI and Anthropic incidents have lent considerable force to the argument that voluntary measures alone may be insufficient.
Bipartisan frustration and the road ahead
Trump’s tech ties have drawn bipartisan fire in the wake of the AI escape incidents. Reuters reported on August 6 that both MAGA-aligned Republicans and Democrats have voiced frustration over what they characterize as inaction in the face of demonstrable risks.
The president’s close relationships with tech industry figures — OpenAI CEO Sam Altman has been a visible presence at the White House — have fueled accusations that the administration is prioritsing industry interests over safety.
Still, the NIST guidelines represent movement on the executive branch side, even if they stop short of binding requirements. The first set of evaluations under the new framework is expected to begin in August 2026, and the public comment period now open will shape the final form the standards take.
As AI capabilities continue to outpace the governance structures meant to contain them, the collision between the voluntary, innovation-forward posture of the White House and the more interventionist instincts on Capitol Hill appears poised to define the next chapter of American technology policy.
Related Posts:
- Meta’s Muse Spark breaches third-party company’s systems
- White House exempts open-weight AI from safety tests
- White House convenes top AI labs for voluntary cybersecurity testing
- Alibaba goes bigger in AI to battle US and Chinese rivals
- Organisations need to boost security architectures to embrace agentic AI