Anthropics Leaks New Model Claude Mythos Amid Cybersecurity Risks and Stock Drop

11 articles | Updated 1d ago | Created 3d ago

China AI

Anthropics recently confirmed via an accidental public cache exposure and subsequent blog post details about its most powerful AI to date named Claude Mythos, which was being tested with early access customers before a major data leak on Thursday exposed the project.

Key Points

1

Anthropics most powerful AI model to date has been identified as Claude Mythos following a data leak of unpublished assets.
2

The leaked draft blog post revealed that the new tier is internally named Capybara and surpasses previous capabilities like Opus, representing 'a step change'.
3

Cybersecurity risks have escalated significantly after reports suggested this model could accelerate AI-driven cyberattacks by exploiting software vulnerabilities faster than before.

Developments

Mar 28, 13:00 Reports confirm the leak of Anthropic's Capybara/Claude Mythos model via unsecured data cache and discuss future implications for cybersecurity.

Mar 27, 14:59 Futurism reports on how a major security lapse in Anthropic's CMS exposed nearly 3,000 unpublished assets including the draft blog post about Claude Mythos.

Mar 27, 18:27 Decrypt notes that cybersecurity stocks declined following reports suggesting the system could accelerate AI-driven cyberattacks and mentions a new tier above Opus called Capybara.

Mar 27, 13:56 Coindesk reports that the leak sent software names and crypto sharply lower due to fears of a cyber arms race.

Here's what next as Anthropic's most powerful AI model leaked via unsecured data cache

Coindesk Shaurya Malwa 1d ago

Anthropics revealed that an unsecured data cache leak exposed details of "Mythos," a new AI model within the Capybara tier described as their most capable system to date and superior in coding, reasoning, and cybersecurity tasks. The company acknowledged this disclosure resulted from human error while confirming they are currently trialing Mythos with early access customers despite flagging unprecedented cyber risks associated with such advanced capabilities.

« L’IA la plus performante que nous ayons jamais créée » : Claude Mythos, le futur modèle d’Anthropic qui menace la cybersécurité

Feeds.leparisien.fr 1d ago

Anthropic redies powerfull Mythos model with high cybersecurity risk

Testingcatalog M1Astra 2d ago

Anthropic inadvertently exposed draft materials revealing that its new AI models—branded as "Claude Mythos" or potentially renamed "Capybara"—represent a significant leap in capability over the current top model (Opus). These leaked documents warn of unprecedented cybersecurity risks associated with this powerful, compute-intensive system.

Anthropic's 'Most Capable' AI Model Claude Mythos Leaks, Deemed Major Cybersecurity Threat

Decrypt Jason Nelson 2d ago

A leaked draft revealed Claude 3 (internally named Mythos), a new tier of AI model larger and more intelligent than Opus with significant advances in reasoning, coding, and cybersecurity capabilities. Cybersecurity stocks declined following reports that the system could accelerate cyberattacks due to its enhanced technical abilities compared to previous models like Opus 4.6 on testing benchmarks for software development and academic tasks.

Leaked Anthropic Model Presents ‘Unprecedented Cybersecurity Risks,’ Much to Pentagon’s Pleasure

Gizmodo AJ Dellinger 2d ago

Fortune reported leaked information revealing Claude Mythos as an unprecedented AI model that surpasses current benchmarks and poses significant cybersecurity risks. While Anthropic confirmed its existence but withheld release due to safety concerns, the Department of Defense remains opposed despite a recent court ruling blocking their attempt to label it dangerous for domestic surveillance or autonomous weapons use.