AI Firms Court Cybersecurity
OpenAI has expanded its 'Trusted Access for Cyber' programme to include large firms such as Bank of America and BlackRock, adding corporate defenders and researchers to the list of trusted partners. Anthropic has also rolled out a new cybersecurity model, which observers say could help its relationship with the U.S. government after earlier friction with the Pentagon. (securitybrief.news; theverge.com)
OpenAI and Anthropic are both steering their newest artificial intelligence tools toward cyber defense, with banks, tech firms, and government evaluators now in the mix. (openai.com; anthropic.com) OpenAI said on April 16 that Bank of America, BlackRock, BNY, Citi, Cisco, Cloudflare, CrowdStrike, Goldman Sachs, JPMorgan Chase, Morgan Stanley, NVIDIA, Oracle, Palo Alto Networks, SpecterOps, Zscaler, and other groups had joined its Trusted Access for Cyber effort. The company said the program gives vetted defenders access to GPT-5.4-Cyber and includes $10 million in application programming interface credits. (openai.com) OpenAI first introduced Trusted Access for Cyber on February 5 as an identity- and trust-based system for handling more capable cyber models, then updated the program on April 14 and expanded the partner list on April 16. It also said the U.S. Center for AI Standards and Innovation and the U.K. AI Security Institute are evaluating the model’s cyber capabilities and safeguards. (openai.com; openai.com; openai.com) Anthropic, meanwhile, launched Project Glasswing on April 7 and said Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks were among its launch partners. Anthropic said the initiative gives defenders early access to Claude Mythos Preview, a general-purpose model it described as its most capable yet for coding and agentic tasks. (anthropic.com) In plain terms, both companies are selling the same basic idea: a model that can read, write, and test code can also help find software flaws before attackers do. Anthropic said Mythos is unusually strong at computer security tasks, while OpenAI said frontier models can reduce response times and help defenders detect and analyze severe attacks. (red.anthropic.com; openai.com) The customer lists show a split in strategy. OpenAI’s April 16 roster leaned heavily toward Wall Street and large enterprises, while Anthropic’s April 7 launch centered more on cloud, chip, software, and infrastructure companies. (openai.com; anthropic.com; bankinfosecurity.asia) Anthropic’s cyber push is also colliding with Washington politics. The Verge reported on April 18 that Anthropic’s Mythos launch came as the company tried to repair ties with the U.S. government after earlier friction with the Pentagon, and Reuters reported on April 17 that Anthropic chief executive Dario Amodei had discussed working together with the Trump administration. (theverge.com; msn.com) That government angle comes with a second argument over risk. Anthropic said it is not releasing Mythos publicly and published a system card and risk report, while OpenAI said higher-capability cyber access should go only to verified users and approved enterprise teams. (anthropic.com; anthropic.com; developers.openai.com; openai.com) The race is no longer just about chatbots for office work. It is now about which AI company can convince banks, security vendors, and governments that its models are useful enough for defense and controlled enough to trust. (openai.com; anthropic.com; theverge.com)
