Shares in major cybersecurity companies dropped sharply on Friday after reports emerged that Anthropic is testing its most capable AI model to date, internally referred to as Mythos, with leaks suggesting the system poses meaningful security risks.
Palo Alto Networks fell approximately 6% and Okta tumbled around 7%, according to Bloomberg and Yahoo Finance data, in moves that extended a difficult stretch for the sector. Cybersecurity stocks have slumped for much of 2026 as investors grow increasingly anxious that frontier AI models will erode demand for conventional security products by automating tasks those platforms are paid to perform.
The immediate concern cuts both ways. A sufficiently capable AI model could lower the cost and skill threshold for mounting cyberattacks, potentially increasing the volume of threats that security vendors must defend against. Meanwhile, investors worry that AI-native tools could disintermediate incumbents by handling threat detection and response without dedicated cybersecurity software.
Analysts, however, caution that the market reaction may be disproportionate. MarketWatch reported that several analysts view AI as ultimately a tailwind for the sector, arguing that more sophisticated threats will drive enterprise spending on security rather than reduce it. That argument has struggled to gain traction with investors this year as model capabilities have advanced faster than expected.
Anthropic has not made a formal announcement about Mythos. The reports appear to be based on a leaked internal post, first surfaced by Fortune and Mashable, describing the model as the most powerful the company has ever developed.
