Cybersecurity Stocks Plunge as Anthropic's Leaked AI Model Raises Unprecedented Security Fears

Anthropic Data Leak Triggers Sharp Decline in Cybersecurity Stocks

Cybersecurity stocks experienced a dramatic selloff on Friday following reports that Anthropic inadvertently exposed internal documents revealing details about a powerful new AI model with advanced cyber capabilities. The Global X Cybersecurity ETF fell as much as 6.1%, bringing its year-to-date decline to more than 20% 1

1

. Market leaders bore the brunt of investor fears surrounding Anthropic AI model, with CrowdStrike and Palo Alto Networks dropping 7%, while Zscaler and SentinelOne tumbled over 8% 2

2

. Tenable plummeted nearly 11%, and both Okta and Netskope fell more than 6% each, illustrating the sector's acute sensitivity to AI-driven disruption 2

2

.

Unprecedented Cybersecurity Risks From Claude Mythos Model

Fortune first reported the breach on Thursday, citing information obtained from a publicly accessible draft blog post that detailed Anthropic's unreleased model, referred to in some documents as Claude Capybara and in others as Mythos 1

1

3

3

. According to the draft blog post, Anthropic believes the model "poses unprecedented cybersecurity risks" 1

1

. The company secured the data on Thursday after being notified, attributing the data leak to "human error in the CMS configuration" 3

3

. An Anthropic spokesperson confirmed the company is testing a model with significantly better performance in reasoning, coding, and cybersecurity, though the error led to the draft being viewable 1

1

.

Concerns About AI Model Capabilities and Hacker Exploitation

The exposed documents describe the new model as a "step change" in capabilities, raising immediate concerns about AI model capabilities being exploited by malicious actors to skirt current cyber defenses 3

3

. Given these advanced cyber capabilities, Anthropic is planning to share results from tests with cyber firms to help them improve their defenses ahead of the model's release 1

1

. The company is testing the model with a small group of early access customers and planning a slow rollout due to potential security risks 2

2

. Bernstein analyst Peter Weed noted that "Anthropic is also trying to limit their product being used by hackers," calling it "good hygiene and a baseline expectation for their product" 1

1

.

AI's Growing Impact on the Sector and Past Incidents

This marks the second time in recent months that Anthropic has triggered selling pressure in cybersecurity equities. Last month, the firm announced a new tool called Claude Code Security designed to scan codebases for software vulnerabilities and suggest targeted patches, sparking a broad drop in cyber names including CrowdStrike and Palo Alto Networks 1

1

3

3

. Earlier this year, a hacker exploited Anthropic's chatbot to carry out attacks against Mexican government agencies, resulting in the theft of sensitive tax and voter information 1

1

. In November 2025, Anthropic confirmed that its Claude Code model was manipulated to execute a cyber-espionage operation across 30 organizations in finance, technology, manufacturing, and government—marking the first confirmed case where an AI agent handled most intrusion steps normally performed by human hackers 4

4

.

Market Reaction and Analyst Perspectives on Buying Opportunity

Despite the sharp selloff, Wall Street analysts quickly defended cybersecurity names. Stephen's Todd Weller said the market is "misinterpreting the news" and called it a buying opportunity for investors 1

1

. Joe Tigay, a portfolio manager at Equity Armor Investments holding Palo Alto, CrowdStrike, and Fortinet across its funds, questioned the logic: "The thing that doesn't add up to me though is that if they are a cybersecurity threat, don't you want the best companies that deal with cybersecurity to be fighting it for you?" 1

1

. The rise of AI and autonomous agents is fundamentally shifting the threat landscape, forcing cybersecurity companies to accelerate innovation to keep pace with more sophisticated attacks and tools that make hacking easier 2

2

. The World Economic Forum's 2025 Cybersecurity and AI report found that 70% of executives said AI has increased their exposure to digital risk, with global cyberattack incidents tripling since 2022 4

4

.