Business

Heavy lifting: Amazon’s Leo LE-01 Mission taking off from the Guiana Space Centre on February 12, with 32 satellites. (Amazon)

Competition between the major LLM makers is alive and well, and May 7 saw the release of Mythos, a new model from Anthropic, which the company said was “strikingly” capable at security tasks. It also launched what it calls Project Glasswing, which it said would help secure the world’s most critical software. Glasswing is invite-only.

Company researchers said Mythos Preview was capable of identifying and exploiting zero-day vulnerabilities in “every major operating system and every major web browser” when asked to do so. A company blog said it had not specifically trained the model to have these capabilities, but they had emerged as a result of improvements in code, reasoning and autonomy. Hot on the heels of Mythos, OpenAI released its GPT-5.4-Cyber, which it said was “cyber-permissive”. It also launched what it calls its Trusted Access for Cyber programme for vetted vendors, researchers and organisations. Anthropic CEO Dario Amodei has met with the White House chief of 4 May 2026 staff as the government seeks access to the model. This is despite the Trump adminstration’s ban on the use of Anthropic’s models after the company disagreed over its use for autonomous weapons and mass surveillance of US citizens. He is also on record as saying that Chinese researchers will be able to duplicate Mythos’ capabilities within six to 12 months.