OpenAI grants EU banks, telcos access to GPT-5.5-Cyber

OpenAI has granted select European banks, telecoms and public services access to GPT-5.5-Cyber through a program called Trusted Access for Cyber. Early participants reported on May 13, 2026 include Deutsche Telekom, BBVA, Telefónica, Sophos, Scalable Capital and the European Commission.

OpenAI says GPT-5.5-Cyber is focused on defensive security tasks. The model can assist teams in analyzing suspicious code, hunting for threats across networks and identifying likely attack vectors. Access is restricted to approved organizations under the Trusted Access for Cyber arrangement.

The company described technical and policy safeguards designed to prevent the model from producing exploit code or other offensive tools. OpenAI said those controls are intended to limit outputs to defensive use cases and reduce the risk that the system could be repurposed for attacks.

Several program participants operate services that touch digital assets. BBVA offers regulated digital-asset services in Europe. Scalable Capital runs investment platforms where tokenized assets are entering regulatory frameworks. Tools that analyze smart contracts and blockchain infrastructure are relevant to those businesses because vulnerabilities in decentralized finance protocols and custody systems can create exposure for custodians and consumers.

Security experts and market participants have cautioned that advanced AI able to review code and detect vulnerabilities can be used for defensive or offensive purposes if misapplied. Observers have raised concerns that gated access to advanced tools may leave smaller crypto firms and startups without comparable defenses, widening the security gap between large incumbents and smaller participants.

OpenAI’s rivals in the enterprise cybersecurity market have been slower to roll out comparable offerings in Europe. Anthropic has not announced equivalent enterprise cybersecurity deployments in the region. OpenAI provided the European Commission access alongside private companies, indicating public-sector defenders are included among approved users.

Stated use cases for GPT-5.5-Cyber include automated code analysis to identify risky patterns, network-wide threat hunting to detect anomalous activity, and mapping likely attack paths to inform human responders.

AI models are increasingly applied to cybersecurity because they can process large volumes of code and telemetry faster than humans alone. The dual-use nature of powerful models has led companies and regulators to consider controls on where and how those systems are deployed. The Trusted Access for Cyber program is an approach that limits distribution while permitting approved organizations to use enhanced defensive capabilities.