Image Credits:Benjamin Girette/Bloomberg / Getty Images4:26 PM PDT · April 21, 2026
A radical of unauthorized users has reportedly gained entree to Mythos, the cybersecurity instrumentality precocious announced by Anthropic.
Much has been made of Mythos and its purported powerfulness — an AI merchandise designed for endeavor information that, successful the incorrect hands, could go a potent hacking tool, according to the company. Now, Bloomberg has reported that a “private online forum,” the members of which person not been publically identified, has managed to summation entree to the instrumentality done a third-party vendor.
“We’re investigating a study claiming unauthorized entree to Claude Mythos Preview done 1 of our third-party vendor environments,” an Anthropic spokesperson told TechCrunch. The institution said that, truthful far, it has recovered nary grounds that the supposedly unauthorized enactment impacted Anthropic’s systems astatine all.
The unauthorized radical tried a fig of antithetic strategies to summation entree to the model, including utilizing “access” enjoyed by the idiosyncratic who was interviewed by Bloomberg. That idiosyncratic is presently employed astatine a third-party contractor that works for Anthropic, the outlet reported.
Members of the radical are portion of a Discord transmission that question retired accusation astir unreleased AI models, the outlet reported. The radical has been utilizing Mythos regularly since gaining entree to it, and provided grounds to Bloomberg successful the signifier of screenshots and a unrecorded objection of the software.
Bloomberg reports that the group, which supposedly gained entree to the instrumentality connected the precise aforesaid time it was publically announced, “made an educated conjecture astir the model’s online determination based connected cognition astir the format Anthropic has utilized for different models.” The radical successful question is “interested successful playing astir with caller models, not wreaking havoc with them,” the root told the outlet.
Mythos was released to a prime fig of vendors, including large names similar Apple, arsenic portion of an inaugural called Project Glasswing. The constricted merchandise of the exemplary was designed to halt its usage by atrocious actors. The instrumentality could beryllium weaponized against firm information alternatively of bolstering it, Anthropic said.
Techcrunch event
San Francisco, CA | October 13-15, 2026
If true, unauthorized usage of Mythos could spell occupation for Anthropic, which provided the exclusive merchandise to allay the company’s interest for endeavor security.
When you acquisition done links successful our articles, we whitethorn gain a tiny commission. This doesn’t impact our editorial independence.
Lucas is simply a elder writer astatine TechCrunch, wherever helium covers artificial intelligence, user tech, and startups. He antecedently covered AI and cybersecurity astatine Gizmodo. You tin interaction Lucas by emailing lucas.ropek@techcrunch.com.















English (US) ·