Anthropic’s ‘Too Dangerous’ Mythos AI Hacked on Day One While White House Plans Rollout
A small group of unauthorized customers accessed Anthropic’s Claude Mythos Preview on the identical day the corporate introduced its managed launch, in accordance with Bloomberg.
The incident raises questions on Anthropic’s potential to comprise a mannequin it deemed too harmful for public launch.
How a Discord Group Walked Into Mythos
Members of a non-public Discord channel devoted to looking unreleased AI fashions made an informed guess in regards to the Mythos endpoint URL.
“Anthropic mentioned Mythos was too harmful to launch. Then 4 random guys in a Discord gained entry on day one by guessing the URL…,” wrote Josh Kale, a preferred consumer on X.
They reconstructed Anthropic’s naming conventions utilizing knowledge uncovered within the Mercor breach three weeks earlier, Bloomberg reported, citing an individual accustomed to the matter.
One group member additionally held reputable analysis credentials by way of contract work for an Anthropic vendor. Those credentials, mixed with the guessed URL, granted the group ongoing entry.
The customers have reportedly been operating Mythos usually since gaining entry. However, they’ve prevented cybersecurity-related prompts and as a substitute used it for benign duties like constructing easy websites.
Anthropic confirmed it’s investigating the report however mentioned it has discovered no proof the entry prolonged past the seller atmosphere.
Anthropic has said Mythos can determine and exploit zero-day vulnerabilities in each main working system and internet browser.
Under Project Glasswing, the corporate restricted entry to roughly 40 accredited organizations, together with Apple, Amazon, and Cisco, strictly for defensive security testing.
White House Pushes Federal Access Despite Pentagon Ban
The breach comes because the White House strikes to develop Mythos entry to civilian federal companies. The Office of Management and Budget emailed Cabinet officers on April 15 outlining plans for a safeguarded version of the model.
This represents a reversal from earlier this 12 months, when the Pentagon designated Anthropic a “supply chain risk” after the corporate refused to take away security guardrails for army use.
“We is not going to let ANY firm dictate the phrases concerning how we make operational choices,” Defense Department spokesman Sean Parnell wrote on X.
A federal decide later paused the broader ban following an Anthropic lawsuit.
Anthropic CEO Dario Amodei met White House officers on April 17, with each side calling talks “productive.”
The NSA has already been utilizing Mythos for vulnerability scanning regardless of the Pentagon blacklist, in accordance with Axios.
The publish Anthropic’s ‘Too Dangerous’ Mythos AI Hacked on Day One While White House Plans Rollout appeared first on BeInCrypto.
