What happened: OpenAI says it’s giving dozens of European firms access to its newest models, including GPT-5.5-Cyber, via a “Trusted Access for Cyber” programme. Names include Deutsche Telekom, BBVA, Telefónica, Sophos and Scalable Capital — basically a who’s-who of “please don’t get hacked”.

Why it matters: OpenAI is trying to keep the sharpest cyber capabilities behind a velvet rope while still giving defenders tools to spot vulnerabilities and respond fast. The promise is access for verified organisations in vital sectors, with safeguards for defensive work.

Wider context: Reuters notes frontier models’ high-end coding ability is raising fears they could be used to find and exploit weaknesses, including in banks and other critical firms. OpenAI explicitly frames this as “trusted defenders” versus everyone else.

Background: OpenAI’s EMEA lead Emmanuel Marill said the goal is balancing access, usefulness and safety as models get more capable. The company has also offered the European Commission open access to cyber features, and said it’s launching a new unit with over US$4 billion in initial investment and buying consulting firm Tomoro.

OpenAI gives European companies access to its latest models to bolster resilience — Reuters

Singularity Soup Take: Congratulations, Europe — you’ve been granted a cybersecurity VIP wristband to the model club, where the bouncers swear they can keep the troublemakers out while still letting the “good guys” borrow the lockpicks.

Key Takeaways:

• Who Gets In: OpenAI says verified organisations in sectors like financial services, telecoms, energy and public services can access its latest models (including GPT-5.5-Cyber), with early names listed such as Deutsche Telekom, BBVA, Telefónica, Sophos and Scalable Capital.
• Safeguards Pitch: The programme is explicitly positioned as defensive: OpenAI says it wants to block dangerous activity while ensuring trusted defenders have tools that are “genuinely useful” for finding vulnerabilities and responding to threats quickly.
• Model Arms Race: The story is framed against a backdrop of rising concern that frontier models can help attackers as well as defenders; Reuters points to rival Anthropic’s recent Mythos release as a signal that cyber risk is becoming a first-class capability category.

Related News

Mythos Makes Bug-Hunting Faster; Patching Still Loses - A recent look at how bug-finding gets cheaper and faster, while the messy human part (actually patching things) stays stubbornly slow.

Mozilla Says Mythos Found Hundreds Of Firefox Bugs - Another datapoint in the “frontier models as vulnerability scanners” trend, with all the fun governance questions about who gets to point them at what.

OpenAI And Anthropic Buy The Consultant Army - On labs building enterprise deployment muscles: product is nice, but an army of implementation humans is how you actually get inside big organisations.

Relevant Resources

ChatGPT: The All-Purpose AI Assistant - A quick refresher on OpenAI’s flagship assistant ecosystem and why “more capable models” inevitably spill into high-stakes domains like security operations.