AI models now weaponize cyber attacks at scale

Key takeaway

• VivaTech highlighted Europe's growing anxiety over AI-driven cybersecurity threats and deepening dependence on U.S. technology providers.

• Advanced AI models can now find unknown vulnerabilities and create working exploits at scale, raising concerns that defenders may fall behind attackers if the window of controlled access closes too quickly.

• Meanwhile, European businesses and policymakers are reckoning with the reality that the U.S. can unilaterally cut off access to frontier AI models, spurring calls for sovereign AI infrastructure—though executives disagree on whether true sovereignty is realistic or pragmatic.

3 Key Points

FAQ

What specific AI capabilities are causing the most concern at VivaTech?

Anthropic's Mythos and OpenAI's GPT-5.5 Cyber can now find unknown vulnerabilities in critical software and generate working exploits at speed and scale—tasks that once required time and specialized human knowledge. The worry is that criminal groups and nation-state actors will gain access before organizations have time to patch their systems.

How are Anthropic and OpenAI trying to manage the cybersecurity risk?

Both companies have staggered the release of their advanced cyber models, granting initial access only to vetted security firms and critical infrastructure operators so they can patch systems before capabilities spread more widely. OpenAI also expanded its Daybreak initiative by pairing GPT-5.5 Cyber with an open-source patching effort called Patch the Planet, working with firms like Cloudflare, Cisco, and CrowdStrike.

What does 'sovereign AI' mean, and why is Europe pushing for it now?

Sovereignty means different things to different actors: some, like Cohere's CEO, call for domestically controlled infrastructure (chips, power, data centers, and private deployment under national control), while others argue for strategic alliances to counter U.S. and China's grip on infrastructure. The U.S. decision to cut off access to Anthropic's frontier models gave the conversation new urgency, making Europe's dependence on U.S.-controlled AI feel much more real.