Anthropic's New Model Tease Spooks Investors As Software Tanks…
Cybersecurity stocks fell Friday after a report said Anthropic had been testing a new artificial intelligence model with stronger cyber capabilities, raising fresh concerns about how AI could reshape the threat environment and pressure security software providers.
The decline spread across the sector. CrowdStrike (NASDAQ: CRWD) and Palo Alto Networks (NASDAQ: PANW) each dropped 6% while Zscaler (NASDAQ:ZS) and SentinelOne (NYSE: S) also lost around 6%. Tenable (NASDAQ: TENB) fell over 9%%, and Okta (NASDAQ: OKTA) and Netskope (NASDAQ: NTSK) each declined 8%.
The move followed a Fortune report that cited a publicly accessible draft blog post and other unpublished materials linked to Anthropic. ResearcFhers who reviewed the exposed cache said it appeared to include nearly 3,000 unpublished assets associated with the company’s blog that had not appeared on Anthropic’s news or research sites.
According to the report, one draft page described a model called Mythos as the most powerful AI system Anthropic has developed so far. The materials indicated the company was taking a cautious approach to release, beginning with a small group of early-access users rather than a broad rollout. The draft also said the model is costly to run and not yet ready for general availability.
Anthropic acknowledged to Fortune that a “human error” in the configuration of its content management system made the draft blog post accessible. The company said the exposed materials were early drafts of content under consideration for publication. After Fortune notified the company of the issue on Thursday, Anthropic removed the public’s ability to search the data store and retrieve documents from it.
In response to questions about the draft materials, Anthropic said it is developing a general-purpose model with advances in reasoning, coding and cybersecurity. A company spokesperson said Anthropic is working with a small group of early-access customers to test the model and described it as a step change and the company’s most capable model to date.
The draft materials also referred to a model tier called Capybara, which Anthropic described as larger, more intelligent and more expensive than its Opus line. The company currently markets Opus as its largest and most capable model family, with Sonnet and Haiku positioned below it in capability and cost. In the unpublished blog text, Anthropic said Capybara posted materially stronger results than Claude Opus 4.6 across tests in software coding, academic reasoning and cybersecurity. Fortune reported that Capybara and Mythos appear to refer to the same underlying model.
The report comes as cybersecurity companies face growing investor concern over the effect of AI on both competition and the threat landscape. More powerful and easier-to-use AI systems have raised fears that cyberattacks could become more sophisticated and simpler to launch, increasing pressure on security vendors to adapt their products more quickly.
The sector has already come under pressure this year amid broader worries about AI disruption in software. Last month, cybersecurity shares also fell after Anthropic introduced a code-scanning security tool for Claude. Anthropic has previously acknowledged that misuse risks exist around its systems. In November, the company said a state-sponsored group in China used Claude to automate a cyberattack, adding to concerns about how advanced AI models could be used in offensive cyber activity.
At the time of publishing, Stocks.News does not hold positions in companies mentioned in the article.