Anthropic and OpenAI invited to NATO summit as Europe pushes back on US AI export controls

AI labs at the alliance table

Anthropic and OpenAI are set to attend NATO’s annual leaders’ summit in Ankara on 7–8 July, according to the Times of India, marking a rare occasion on which two privately controlled frontier artificial-intelligence developers are formally embedded in the alliance’s programme. The summit, hosted by Turkish President Recep Tayyip Erdoğan and attended by US President Donald Trump, is the first NATO gathering of Trump’s second term and comes against a backdrop of reopened tensions between Washington and European capitals over technology, defence spending and the war in Ukraine.

Why the two US labs matter to NATO

According to the Times of India, Anthropic and OpenAI have in recent weeks unveiled new models that they say can find and exploit software vulnerabilities at a level beyond most human engineers. The report cited Anthropic’s model ‘Claude Mythos’, which it said surfaced vulnerabilities in classified US systems within hours during a government test. The combination of offensive and defensive applications was significant enough to prompt a rare joint warning from members of the Five Eyes intelligence-sharing alliance — the United States, United Kingdom, Canada, Australia and New Zealand — urging global leaders to ‘swiftly’ step up security against AI-powered cyber threats.

That capability is exactly what makes the models strategically sensitive. NATO members see potential both for hardening cyber defences and for adversaries using the same tooling to mount attacks at unprecedented speed. According to the Times of India, that dual-use profile is what has put Anthropic and OpenAI at the centre of this year’s summit agenda, alongside conventional items such as burden-sharing, the situation in Ukraine and Türkiye’s role as host.

Washington’s export controls and Europe’s frustration

The Trump administration has used that sensitivity to keep tight control over who can use the tools. In early June the administration imposed export controls on Anthropic’s most cyber-capable models, ‘Mythos’ and ‘Fable’, banning both foreign nationals and foreign countries from accessing them, the Times of India reported. The White House also intervened to limit the rollout of OpenAI’s latest cutting-edge model to a small group of US companies it had pre-approved.

European governments have pushed back, the report said. Germany, among other states, asked for access to Claude Mythos after its April announcement, but was initially denied. The United Kingdom was among the first non-US countries allowed to test the technology, and earlier this month Anthropic expanded its testing network to around 150 organisations across 15 countries, including European Union members. European officials remain uneasy, the Times of India reported, citing what it described as a persistent ‘push-and-pull’ tactic and the precedent set by US software vendors of invoking a ‘kill switch’ to disable systems abroad.

Europe’s parallel track: Helsing and Mistral

Frustrated by the bottleneck, European technology companies have spent the past year building a parallel capability. The Times of India reported that German defence-tech firm Helsing, founded in 2021, and French AI developer Mistral — described by the paper as Europe’s closest competitor to OpenAI — formed a partnership last year to integrate AI models into European defence platforms. Helsing already supplies software deployed in battlefield simulations, fighter jets and drones used in Ukraine, according to the report. Mistral has meanwhile drawn strong investor demand across Europe in recent months. The two companies say their collaboration will support decision-making, environmental awareness and operator-machine communication in military settings.

The Times of India noted that the partnership was driven in part by anxieties over Russia’s posture toward its neighbours and by the Trump administration’s direct pressure on NATO spending baselines, both factors that have propelled European military budgets upward over the past year.

Defence money vs. defence delivery

The AI row is happening as Europe struggles to translate political pledges into weapons, factories and budgets. The Economist reported from Örnsköldsvik in northern Sweden that BAE Systems-owned Hägglunds, the manufacturer of the CV90 infantry fighting vehicle, has more than tripled its workforce to about 2,600 since 2021, with revenue jumping from $211 million in 2018 to $1.1 billion in 2025 and a forecast of more than $2 billion. A Ukrainian flag presented by the 21st Brigade — which operates the CV90 and calls it ‘the Scandinavian beast’ — now hangs in the facility. But The Economist, writing under the headline ‘Europe promised cash for defence. It’s failing to cough up’, stressed that ‘most Europeans want stronger armed forces, but are unwilling to pay for them’, a warning that AI capability will not fix underlying shortfalls in industrial output and public funding.

The Washington Post framed the Ankara meeting as a direct test of Trump’s 2025 spending victories inside the alliance, with the paper noting that Trump ‘won big spending promises from NATO last year’ and would ‘try to enforce them’ in Ankara. The Post also flagged Iran policy and Erdoğan as further pressure points on the summit agenda.

A wider rupture inside the alliance

The Wall Street Journal reported that European leaders held an emergency January meeting to discuss managing a possible further rupture with Washington after Trump’s early second-term actions, with the NATO Secretary-General attending. The piece, whose headline reads ‘There Is No Going Back: The Inside Story of Europe’s Rupture With America’, signals that the AI export dispute sits inside a broader transatlantic split the alliance is now trying to manage.

JPMorgan chief executive Jamie Dimon added a sharper economic edge to the picture. Speaking at the Council on Foreign Relations’ CEO Speaker Series and quoted by the Times of India via Moneywide, Dimon called Canadian Prime Minister Mark Carney’s call for middle powers to form stronger coalitions ‘a fantasy’, saying: ‘They did that. It’s called Europe.’ Dimon said Europe’s GDP has slipped from 90 percent of America’s to 70 percent and attributed that gap to high taxes, regulation and weak capital markets, urging Europe to build ‘a true common market with open trade in services’ rather than rely on geopolitical blocs.

What to watch at Ankara

For the AI track specifically, three concrete signals will indicate whether the summit changes anything. First, whether NATO publishes any communique language acknowledging AI export controls or calling for allied access to US frontier models. Second, whether Anthropic expands its 15-country, 150-organisation testing programme during or immediately after the summit, given that some EU members were already inside that cohort. Third, whether Helsing and Mistral, which have been quietly building the European alternative, announce a new platform deployment before the alliance’s autumn ministerial.

In the background, the underlying facts the Times of India laid out are unchanged as of publication: Anthropic and OpenAI remain US-controlled firms operating under White House export restrictions; European militaries remain dependent on Washington for the most capable cyber AI; and the Helsing-Mistral partnership remains the most credible indigenous alternative, without yet matching the scale of the US labs.