Washington is building a two-track AI order: confidential security reviews for powerful closed models, regulatory relief for open-weight systems, and a G20 campaign for light-touch rules. The strategy may accelerate US innovation, but leaves a critical security gamble unresolved for global markets.
Innovation Intelligence tracks the technologies, companies and capital flows shaping the next industrial phase of AI. This week's second story covers a critical policy decision in Washington, as the US government attempts to balance national security against technological innovation by formally dividing the regulatory landscape between closed and open-source artificial intelligence.
The White House has finalised its highly anticipated framework for evaluating the security risks of advanced artificial intelligence, drawing a hard regulatory line between proprietary and open-source technology. The Trump administration's new guidelines establish a voluntary, pre-release government review process focused exclusively on closed, proprietary AI models that demonstrate state-of-the-art capabilities in cybersecurity and hacking. Crucially, open-weight models—systems whose underlying code is made publicly available by developers like Meta and NVIDIA—are explicitly exempt from the government vetting process.
This policy decision reflects Washington's strategic gamble: attempting to secure the most dangerous frontier models without stifling the broader American AI ecosystem. The framework is designed to force companies developing closed systems, primarily OpenAI, Anthropic, and Google, to submit their models to federal security agencies for up to 30 days of confidential benchmarking before public release.
While technically voluntary, the political and legal risks of bypassing the White House review are severe; OpenAI recently delayed a product launch to accommodate the framework's finalisation, and Anthropic previously pulled models after government security flags.
The exemption for open-source AI is a massive victory for companies like Meta, which have argued that open development accelerates innovation and democratises access. However, the framework's reliance on ambiguous definitions of "state-of-the-art" cyber capabilities leaves significant grey areas. As the White House prepares to brief industry executives, the message is clear: the US government intends to heavily scrutinise the black boxes of the AI elite, while giving open-source developers a regulatory free pass to accelerate American technological dominance.
That domestic division now sits inside a much broader American campaign to shape international AI policy.
At the G20 Innovation Ministerial in Chapel Hill, North Carolina, the United States secured support for the non-binding Carolina Principles for Emerging Technology. The joint position urges governments to rely on existing laws wherever possible and introduce AI-specific rules only when the technology creates genuinely “novel considerations”. China, despite already operating an extensive AI regulatory system, was among the countries supporting the statement.
Meta chief executive Mark Zuckerberg argued against restrictions on open-weight models, while NVIDIA chief executive Jensen Huang urged governments to regulate demonstrated, real-world harms rather than theoretical risks. Google DeepMind’s Demis Hassabis offered a more cautious counterpoint, calling for internationally coordinated safety testing.
Washington also pushed the G20 towards a permissive approach to training data. Commerce Secretary Howard Lutnick called for frameworks that protect creators while allowing AI developers to use copyrighted material under fair-use principles. On the same day, the US Justice Department filed a brief supporting OpenAI in its copyright dispute with The New York Times, arguing that AI training generally constitutes transformative fair use.
Washington is constructing a two-track AI order. At home, it reserves confidential security scrutiny for the most capable closed models while giving American open-weight developers room to scale. Internationally, it is pressing governments to avoid new regulators, preserve access to training data and keep open-model distribution relatively unrestricted.
That combination strengthens Meta and other open-model developers, but the burden on OpenAI, Anthropic and Google should not be overstated. The review remains voluntary and limited to models crossing a classified cyber-capability threshold. It is not a universal 30-day approval process.
The larger strategic calculation is clear. The Trump administration appears willing to accept greater diffusion risk from open models if doing so prevents Chinese systems from becoming the default foundation of the global AI economy. Security reviews are being concentrated around the most powerful proprietary black boxes, while open-weight AI is being treated as infrastructure for American technological influence.
It is a calculated gamble, not a regulatory free pass. The unanswered question is whether the next serious AI-enabled cyber incident emerges from the closely supervised frontier laboratories, or from the open ecosystem Washington has chosen to accelerate.
Get the stories that matter to you.
Subscribe to Cyber News Centre and update your preferences to follow our Daily 4min Cyber Update, Innovative AI Startups, The AI Diplomat series, or the main Cyber News Centre newsletter — featuring in-depth analysis on major cyber incidents, tech breakthroughs, global policy, and AI developments.
Washington handed AI’s builders medals while the bond market sent the bill. With 30-year yields at their highest since 2002 and Firmus shelving its float, the question is no longer whether the AI race is real, but who can still carry the debt when the cost of money finally catches up with ambition.
Beam and Mistral are challenging China’s lead in open-weight AI, promising far greater efficiency, choice and sovereign control. For Australia and the Global South, the opportunity is to host intelligence on their own terms, even as unverified benchmarks and cyber risks demand closer scrutiny.
OpenAI’s widening agent crisis is testing confidence in autonomous AI, as Wikimedia reports unauthorised activity and Australia seeks answers over government system breaches. Beyond stolen data, the stakes include disrupted services, public trust and who pays when AI acts beyond its own authority.
An OpenAI agent breached a Medicare portal, and Canberra waited 83 days to find out. Now Albanese is weighing prosecution, the Senate has summoned Altman, and OpenAI admits dozens more were hit. If Australia can be kept in the dark this long, what chance do nations that cannot ring Sam Altman have?
Where cybersecurity meets innovation, the CNC team delivers AI and tech breakthroughs for our digital future. We analyze incidents, data, and insights to keep you informed, secure, and ahead. Sign up for free!
