Key Takeaways

  • Palmer Luckey warned that outdated systems are struggling against attackers using AI to discover and exploit vulnerabilities faster.
  • Survey data shows a wide gap between reported AI-powered attacks and the deployment of AI-enabled defenses.
  • Organizations can narrow that gap through automated detection, stronger AI oversight, modernized systems, and regular red-team exercises.

Palmer Luckey delivered a blunt assessment of the cybersecurity landscape at The New York Times DealBook Summit on Wednesday, arguing that attackers are adopting artificial intelligence faster than many organizations can update their defenses.

The Oculus founder and Anduril Industries co-founder connected the recent spate of technology outages to a broader mismatch. Attackers use advanced AI tools to find vulnerabilities, develop exploits, and adjust campaigns quickly, while defenders often depend on older systems not designed for that pace.

“We’re seeing the natural effects of a lot of these advanced AI tools allowing for the creation of various exploits,” Luckey said.

Government officials have largely attributed the latest technical problems to companies rushing to patch systems against newly discovered vulnerabilities. Luckey focused further upstream on the aging infrastructure and slow operational processes that make emergency patching so disruptive.

Installing another security product does not automatically fix brittle applications, incomplete asset inventories, delayed patch cycles, or fragmented identity controls. AI accelerates detection, but it can also amplify the consequences of weak security foundations.

Attackers do not face the same procurement cycles, governance reviews, integration projects, and staffing constraints as large enterprises. They can test an AI-assisted technique, discard it, and try another rapidly, outpacing the speed at which corporate security teams can change their own operating models.

Industry data underscores the widening gap. BCG reported in 2025 that 60% of surveyed companies believed they had experienced an AI-powered attack during the prior year, yet only 7% had deployed AI-enabled defenses. Although 88% planned to deploy them, planned investment offers little protection against attacks already underway.

The budget picture is similarly uneven. BCG found that 53% of executives ranked AI-related cyber threats among their top three organizational risks, while only 5% reported substantially increasing cybersecurity budgets because of AI attacks. Meanwhile, 69% of organizations had difficulty hiring AI-cybersecurity talent, and just 25% considered their existing AI-enabled defenses advanced.

Threat expectations are spreading across several categories. Financial fraud led at 43%, followed by social engineering at 39%, AI-assisted vulnerability discovery at 28%, and adaptive malware at 26%. Because AI helps attackers personalize lures, analyze exposed systems, and revise malicious code with less manual effort, the attacks can scale rapidly.

Preparation appears inconsistent inside enterprise AI programs. In a global survey of more than 3,000 defenders, Trend Micro found that only 6% regularly conducted AI audits or red-team exercises. Security teams were involved during AI discovery at 23% of organizations and during pilots at 25%. By the time security teams evaluate post-deployment environments, architectural choices and data flows are often difficult to change.

Luckey also placed the issue in a national-security context, stating that U.S. adversaries exploit the gap for strategic and financial purposes, singling out North Korea as particularly involved. Andrew Ross Sorkin noted that Vice President J.D. Vance had canceled his DealBook Summit appearance because the recent outages required him to attend meetings on the matter.

Asked how worried people should be, Luckey replied, “It depends on how well prepared you are.” He added, “You guys are going to find out what’s going on soon enough.” No detailed technical explanation of that warning was provided at the event.

Despite the immediate risks, Luckey argued that AI could eventually favor defenders, provided they adopt it effectively. Automated defense can correlate signals across endpoints, identities, cloud environments, and network activity faster than a human analyst working through separate consoles.

Companies such as CrowdStrike illustrate the push toward automated detection, while Palantir and Anduril reflect a wider move toward AI-assisted decision systems in security and defense. The competition increasingly centers on autonomous or semi-autonomous operations on both sides.

NIST has offered organizations a practical starting point through the NIST Cybersecurity Framework 2.0 and its Cyber AI Profile, IR 8596. The profile organizes the challenge around securing AI systems, using AI for cyber defense, and thwarting AI-enabled attacks.

For business leaders, moving faster does not mean deploying untested automation across sensitive infrastructure. It requires shortening detection and response cycles, involving security earlier in AI projects, testing models and connected systems, and replacing unsupported technology in manageable stages. Luckey’s final message captured the urgency: “It’s time to start running.”