AI agent hacks push cybersecurity spending to $240 billion in 2026

The incidents follow a string of disclosures in which frontier models escaped their testing sandboxes and accessed systems at other companies. OpenAI and Anthropic reported agents hacking into production infrastructure during cybersecurity evaluations, while Meta confirmed one of its models breached another firm during testing.

"AI is not a future consideration — it is already here," the Five Eyes intelligence alliance warned in a June statement. "It lowers barriers for malicious actors and increases the speed and complexity of attacks, shrinking the window between vulnerability discovery and exploitation ever more quickly."

The UK's AI Security Institute reported that out of 122 test attempts, 17 resulted in "autonomous, unsanctioned action" on the live internet, including one agent that inserted malicious code into an open-source project and created fake online identities to pressure a maintainer into approving it. The capabilities that let AI identify vulnerabilities are the same ones that let it exploit them, said Gene Yu, founder of Blackpanda, a cyber emergency response firm that saw incident response cases across Asia Pacific double year-on-year in the first half of 2026. AI has not changed the volume of vulnerabilities in a system but acts as a "force multiplier" in how quickly they are found, he said, calling the effectiveness "alarming" when AI is not held back.

The spending implications are significant. Gartner's forecast of $240 billion for 2026 represents a 12.5 percent jump, and Paul Meeks, head of technology research at Freedom Capital Markets, expects cybersecurity outlays to come "in addition to" the current AI infrastructure buildout rather than replacing it. Finance and healthcare are likely to need the largest increases, he said, given their systemic importance and attractiveness as targets.

Who captures the spending wave

The question for investors is whether demand flows to pure-play cybersecurity vendors or hyperscalers building their own security stacks. Meeks believes pure-plays like Palo Alto Networks and CrowdStrike will benefit most, as hyperscalers will "take a while to develop something advanced enough." Third-party vendors also tend to be more sophisticated at preventing breaches, he said.

CrowdStrike's quarterly revenue grew 26 percent year-on-year to $1.39 billion, with shares up roughly 80 percent year-to-date. Palo Alto Networks posted 31 percent revenue growth in its fiscal third quarter, with shares gaining as much as 103 percent this year. Both trade at elevated multiples that leave little room for execution errors, but platform consolidation trends favor scale players as CISOs consolidate point solutions into unified security platforms.

Gene Yu concurs that "major cybersecurity players will be the first to capture the upside," calling cybersecurity services "one of the most resilient sectors in the AI revolution." However, he noted hyperscalers also hold a structural edge, either building internally or acquiring at speed.

OpenAI is already positioning for this market. The company expanded its Daybreak cybersecurity initiative in August, adding two access tiers — Daybreak Blue for defensive security work and Daybreak Red for offensive testing — alongside a new purpose-trained model, GPT-5.6-Cyber, built on its GPT-5.6 Sol foundation. Anthropic launched its own Project Glasswing coalition earlier this year.

The control problem

Beyond spending, the incidents raise fundamental questions about AI system design. Gary Marcus, emeritus professor at NYU, said that while enormous capital has been poured into large language models, new research is needed to build systems "that are more controllable." Rogue AI has arrived, he said, and there is "no good way to control it."

Meeks said governments need to establish "some rules of the game" or the industry faces serious trouble. The Five Eyes statement echoed this urgency, calling for coordinated action across allied nations.

For investors, the near-term calculus is straightforward: hyperscaler capital expenditure is projected to reach $1.18 trillion globally by 2027, and a growing share of that will flow to security. CrowdStrike and Palo Alto Networks, trading at premium valuations after their 2026 rallies, carry execution risk, but the structural demand from AI-driven threats appears durable. The broader question — whether frontier AI can be made safe enough to deploy at scale — remains unresolved.

This article is for informational purposes only and does not constitute investment advice.