Cybersecurity stocks slide after report says Anthropic is testing a more powerful AI model

Cybersecurity stocks fell sharply on Friday after reports said Anthropic is testing a new advanced AI model with stronger coding and cybersecurity abilities than its previous top system. The sell-off hit several major names across the sector and pushed the Global X Cybersecurity ETF to its lowest close since November 2023.

The market reaction followed a Fortune report on a leaked Anthropic draft blog post that described a new model tier called Capybara and a flagship model called Mythos. According to that report, Anthropic said the system scored much higher than Claude Opus 4.6 on software coding, academic reasoning, and cybersecurity tests, and the company limited access to a small group of vetted early users.

Investors appeared to read the report in two ways at once. First, a stronger AI model could help automate vulnerability discovery and speed up security work. Second, the same capability could raise fears that attackers may also use similar systems to find flaws faster, which would put more pressure on traditional security vendors. Barron’s and Investopedia both said those concerns helped drive Friday’s move lower across the cyber sector.

What happened in the market

The biggest listed cybersecurity names all lost ground during the session. Bloomberg reported that CrowdStrike, Palo Alto Networks, and Zscaler each fell more than 5%, while Cloudflare dropped 3.4%. The Global X Cybersecurity ETF fell 4.5% on the day and deepened its 2026 decline to more than 21%.

The weakness did not happen in isolation. The broader U.S. market also sold off Friday as geopolitical tensions and weaker consumer sentiment weighed on stocks, which likely added to the pressure on high-growth software names. That means Anthropic-related concerns were a major catalyst for cybersecurity shares, but not the only force dragging them lower.

Even so, market coverage tied the cyber drop directly to the Anthropic report. MarketWatch, Barron’s, and Investor’s Business Daily all said investors reacted to fears that more powerful AI systems may change the competitive balance in cybersecurity and raise the pace of AI-driven attacks.

Why Anthropic’s reported model drew so much attention

The reported internal draft described Mythos as Anthropic’s most powerful model yet. Fortune said the document called it a “step change” in capabilities and said the company was moving slowly because of the system’s unusually strong performance and the risks tied to advanced cyber abilities.

That matters because cybersecurity models sit in a dual-use category. A system that helps defenders discover bugs, trace code paths, or analyze complex software can also help attackers identify exploitable weaknesses. Barron’s said the fear was not simply that AI would replace security products, but that stronger AI could overwhelm existing defenses unless vendors adapt quickly.

Anthropic already publishes broader safety and risk materials for advanced models, including its responsible scaling roadmap and periodic risk reports. Those documents show the company has an existing framework for handling dangerous capability thresholds, though the specific Mythos details in this case came from leaked draft material reported by Fortune rather than a formal public launch page.

Market move at a glance

Source basis: Bloomberg and follow-on market coverage.

What this could mean for cybersecurity vendors

Friday’s sell-off reflected fear, but not a settled verdict. Some analysts quoted in market coverage argued that stronger AI may end up helping cybersecurity companies that build AI-native defenses, rather than destroying the sector outright. MarketWatch noted that Palo Alto Networks CEO Nikesh Arora bought about $10 million in company stock during the dip, which signaled confidence that the market reaction may have overshot near-term reality.

That view has some logic behind it. If advanced models make vulnerability discovery faster, companies may need more security tooling, not less. The pressure would likely fall hardest on vendors that depend on slower, rules-heavy approaches and less on firms that can absorb AI into detection, response, and threat-hunting workflows. This is an inference based on the analyst commentary and sector reaction, not a confirmed outcome.

The bigger issue behind the stock move

The real concern is not one leaked model name. It is the growing belief that AI systems are moving from assistant tools into full-spectrum cyber operators that can reason through code, surface vulnerabilities, and accelerate offensive and defensive work alike. Investopedia said the Friday drop deepened an already weak year for cybersecurity stocks, which suggests investors now see AI disruption as a structural issue, not a one-day headline.

That does not mean current cybersecurity companies are obsolete. It does mean investors now expect them to prove they can keep up with faster model-driven threat analysis and response. The companies that show clear AI execution may recover faster than the ones that look exposed. That final point is an inference from the market reaction and analyst commentary.

FAQ