On August 4, a group of companies, reportedly including Anthropic, Google, OpenAI and Meta; met with the US Office of the National Cyber Director to be briefed on the latest draft of a policy framework to be used in reviewing artificial intelligence models used by the government.
The framework is to be used by AI developers to review their platforms for risks that they could be used by attackers to undermine cybersecurity measures, the protection of intellectual property and confidentiality. It also will place parameters around the release of new models for use by the public and for export.
Sources reported that the framework, which would give the US government access to models for up to 30 days before they are released, is voluntary in nature and may be kept confidential. Being voluntary, it apparently will not have the force of law.
Why it matters
The potential for artificial intelligence platform to serve both as a way to detect and fight piracy, and as a weapon to conduct massive copyright infringement have become well documented – with the latter having been the subject of lawsuits by online publishers and the media and entertainment industry.
In recent years, the US government has embraced AI and the companies developing AI platforms, have also increasingly used AI in military applications. In July 2025, the US Department of Defense (aka ‘War’) awarded a $200 million contract to Anthropic, Google, Musk-owned xAI and OpenAI to “accelerate DoD adoption of advanced AI capabilities to address critical national security challenges,” for “lawful use.” [ see Note below ].
Objections have been raised about the weaponization of AI platforms by the US government, including disinformation campaigns, the tracking of US persons within the United States and violations of rights under the US Constitution.
While the ‘lawful use-cases‘ were not named explicitly by the government, Anthropic objected to two possible uses of its Claude technology: to conduct mass surveillance within the United States and to use it in autonomous weapons applications.
In February 2026, the Trump administration designated Anthropic to be a supply chain risk and barred Anthropic from selling into the US government; which some characterized as a revenge move against Anthropic’s stated position. Trump said that he “fired them like dogs…”
And yet, Anthropic has the implicit blessing of the US government, as participants in, and contributors to, this AI platform benchmarking process and AI model framework.
Further reading
White House finalises voluntary AI hacking tests for frontier models. Article. August 4, 2026. Civil Society Alliances for Digital Empowerment (CADE, co-funded by the European Union).
White House finalizes artificial intelligence oversight framework. Article. August 3, 2026. by Cheyenne Haslett and Sophia Cai. Politico
OpenAI CEO Sam Altman heads to Washington as AI policy deadline nears. Article. July 27, 2026. by Cheyenne Haslett and Sophia Cai. Politico
Trump executive order seeks pre-release access and assessment of AI models. Article. June 9, 2026. Swarmnetics
How AI firm Anthropic wound up in the Pentagon’s crosshairs. Article. March 9, 2026. by Nick Robins-Early. The Guardian
CDAO partners with frontier AI companies. Article. July 15, 2025. by Loren Blinde. Intelligence Community News.
—
[ Note: the US Chief Digital and Artificial Intelligence Office (CDAO) within the US Department of Defense still lists its July 2025 announcement among its press releases, but it was no longer clickable as of August 2026. ]
