OpenAI adds Telefonica and Deutsche Telekom to TAC programme

TL;DR

• OpenAI added Telefonica and Deutsche Telekom to its Trusted Access for Cyber programme, according to Reuters.
• The programme gives vetted organisations access to GPT-5.5 and, for a smaller group, GPT-5.5-Cyber for defensive cybersecurity work.
• Reuters also reported that Sophos, BBVA and Scalable Capital were included, and that the European Commission was offered access to the latest cybersecurity models.

OpenAI added Telefonica and Deutsche Telekom to its Trusted Access for Cyber programme, according to a Reuters report cited by Mobile World Live. Reuters said the programme also includes Sophos, Spain-based BBVA and UK-based Scalable Capital, and gives vetted organisations access to GPT-5.5 and OpenAI's “most permissive” GPT-5.5-Cyber model.

OpenAI's Trusted Access for Cyber programme gives approved organisations broader access to cybersecurity model capabilities than standard commercial models for defensive security work. The report said the programme keeps safeguards in place to block misuse including credential theft, malware deployment and exploitation of third-party systems.

Under the programme, approved organisations receive lower refusal rates on cybersecurity requests tied to authorised workflows including vulnerability identification, malware analysis, binary reverse engineering and secure code reviews. OpenAI positioned GPT-5.5 for most defensive use cases, while GPT-5.5-Cyber is available to a smaller group handling specialised tasks such as authorised penetration testing and red teaming.

Earlier in the week, the European Commission confirmed that OpenAI had offered the regulator access to its latest cybersecurity models. Mobile World Live also reported that concerns about AI-enabled cyber threats have increased following the release of Anthropic's Mythos model last month.

Related questions

• What is OpenAI's Trusted Access for Cyber programme?
• Which companies were added to OpenAI's TAC programme?
• What is the difference between GPT-5.5 and GPT-5.5-Cyber in cybersecurity use cases?