← Back to feed

OpenAI, Anthropic and Alphabet’s Google announced on April 6, 2026 that they are sharing intelligence on Chinese industrial espionage and adversarial model distillation through the Frontier Model Forum — marking the first time the top three US frontier labs have formally aligned on a single adversary, and the first time the FMF has been used as an operational threat-intelligence clearinghouse rather than a policy-and-papers organisation.

What They Committed To

The joint statement, signed by Sam Altman, Dario Amodei and Demis Hassabis, commits the three labs to four concrete obligations:

• Real-time threat sharing. When one lab detects a credible espionage or distillation attempt — phishing, insider recruitment, supply-chain compromise, or unauthorised weight extraction — it must notify the other signatories within 48 hours through a secured FMF channel.
• Joint forensic analysis. Shared incident data is pooled and analysed by a rotating technical working group drawing staff from all three labs, enabling correlation of tradecraft, tooling and attribution indicators across targets.
• Public disclosure. The FMF will publish quarterly threat reports naming techniques and, where legally possible, attributed actors. The first report is scheduled for July 2026.
• Coordinated US government response. The pact formalises a single point of contact with the FBI, CISA and other federal agencies for espionage and distillation incidents.

Why Now

Three forces converged in Q1 2026. First, confirmed espionage incidents — including the 2024 arrest of former Google engineer Linwei Ding for IP theft — gave the abstract risk a face. Second, Anthropic publicly identified three Chinese AI labs — DeepSeek, Moonshot and MiniMax — as engaged in illicit distillation from Claude, and OpenAI sent a memo to Congress in February accusing DeepSeek of using “increasingly sophisticated tactics” to extract results from its models. Third, the economic stakes crossed a threshold: US officials estimate unauthorised distillation costs Silicon Valley AI companies roughly $500 million annually in foregone competitive advantage.

The distillation threat is distinct from traditional IP theft. Adversarial distillation uses a target model’s API outputs as training signal to build a functionally similar “student” model at a fraction of the original development cost. The resulting models often lack the safety guardrails embedded in the source model — a national security concern as much as a commercial one.

The Microsoft Absence

Microsoft, the fourth FMF founding member, is not a signatory. Its relationship with OpenAI — which includes equity stakes and deeply integrated Azure infrastructure — creates antitrust complexity around joint threat-response commitments. The FMF has indicated that a governance charter addressing these ambiguities will be published before the July 2026 threat report.

For now, information sharing on distillation remains limited by AI companies’ uncertainty about what is permissible under existing antitrust guidance. The firms would benefit from greater clarity from the US government, according to sources familiar with the matter.

Signal

The intelligence-sharing pact is small in scope — limited to detecting and documenting adversarial distillation — but large in what it signals: the top three US AI labs have concluded that Chinese espionage risk is significant enough to overcome the competitive instinct to keep security information proprietary. Google already published a blog documenting an increase in model-extraction attempts; Anthropic blocked Chinese-controlled companies from Claude access last year.

The FMF was a policy-and-papers organisation for its first three years. The April 6 pact is its first operational mandate, and it names a specific adversary.