OpenAI Buys Its Way Into Europe While Anthropic Waits
What happened
The European Commission announced Monday that OpenAI has offered EU cybersecurity defenders open access to its GPT-5.5-Cyber model through an 'OpenAI EU Cyber Action Plan,' personally championed by former British Chancellor George Osborne. Rival Anthropic has had four to five meetings with Commission officials but has not offered comparable access to its advanced Mythos model. The offer came one month after the EU Commission classified ChatGPT as a large online search engine subject to the Digital Services Act, creating significant regulatory exposure for OpenAI in Europe.
OpenAI is not giving Europe a security tool. It is buying regulatory goodwill by trading access for leniency, one month after the EU moved to classify it as a DSA-regulated platform.
The Hidden Bet
The offer is about cybersecurity
The timing is too precise to be coincidental. The offer followed DSA classification by exactly one month. What the EU gets is a tool. What OpenAI gets is a cooperating posture in ongoing enforcement proceedings worth potentially billions in fines.
Anthropic is simply slower or less cooperative
Anthropic may be calculating that concessions on model access set a precedent it cannot afford. If you give governments access to Mythos now, they will expect access to whatever comes next, with no corresponding reduction in regulatory scrutiny.
EU enforcement of the DSA will be impartial
Regulators are human. A company that cooperates proactively and publicly, with a named ex-finance minister as its representative, is harder to fine in the same press cycle. The offer creates political friction for any enforcement action.
The Real Disagreement
The fork: should AI companies buy regulatory peace through selective cooperation, or hold the line and force governments to either regulate consistently or not at all? OpenAI's bet is that selective cooperation extends its operating window in Europe and sets terms favorable to incumbents. Anthropic's bet is that any precedent of government model access becomes a floor, not a ceiling, and the erosion never stops. Both are right about the dynamic. The disagreement is which risk is worse. I lean toward Anthropic's position being more coherent long-term, but OpenAI's position being more survivable short-term.
What No One Is Saying
The EU just received an offer from a US company to give European government security agencies access to advanced AI capabilities that the same company refuses to open-source or explain. The 'democratic access' framing in OpenAI's letter is covering for a deal that hands a small group of accredited European institutions exclusive access unavailable to researchers, civil society, or the public.
Who Pays
Anthropic
Over the next 12 months as DSA enforcement proceedings advance
If OpenAI's cooperative posture softens EU enforcement against them specifically, Anthropic faces comparatively harsher treatment for identical conduct, giving OpenAI a structural regulatory advantage in Europe.
European researchers and civil society
Immediately, as access is granted
Government security agencies get model access. Universities, journalists, and privacy advocates do not. The same AI that will monitor European citizens is shared with the state, not with those who monitor the state.
Scenarios
Deal holds, Anthropic bends
OpenAI secures favorable DSA treatment and Anthropic, facing comparative disadvantage, offers similar access within six months. The precedent becomes industry norm: share models with governments to avoid regulation.
Signal Anthropic announces an EU-facing model access initiative before September 2026
Deal holds, Anthropic holds
OpenAI gets regulatory breathing room in Europe; Anthropic faces a series of enforcement actions that OpenAI escapes. Over 18 months, OpenAI's European market share expands as Anthropic's operational costs in Europe climb.
Signal EU Commission opens a formal DSA investigation against Anthropic but not OpenAI in the same period
Backlash resets the deal
Privacy advocates and academic AI researchers publicly frame the deal as regulatory capture. European Parliament members raise questions. The Commission walks back the arrangement or makes it formally public with oversight conditions attached.
Signal European Data Protection Board issues an opinion on the access arrangement
What Would Change This
If it emerged that OpenAI's GPT-5.5-Cyber offer was explicitly conditioned on DSA lenience in private negotiations, the framing shifts from 'cooperative' to 'quid pro quo' and the Commission faces a legitimacy crisis.