The Board Room

Deep dives

1. 01

   AI Finds Your Vulnerabilities Before You Do — And Your AI Tools Are Creating New Ones

   The Offense-Defense Equation Just Broke

   When Wiz CTO Ami Luttwak — now operating under Google's umbrella — says the new AI models are 'essentially the best cybersecurity researchers in the world, and that's a problem,' he's describing a threshold that ten independent sources this cycle confirm has been crossed. Anthropic's upcoming model found 500+ high-severity vulnerabilities in mature open-source software including Ghost CMS, the Linux kernel, Vim, and Emacs — using prompts as simple as 'find a vulnerability.' One critical Ghost vulnerability had been missed by 13 years of human security research.

   The cost of vulnerability discovery just collapsed to near-zero. Your patch velocity, risk scoring, and security SLAs were all calibrated for a world where finding zero-days was expensive and slow. That world ended this week.

   Your AI Tools Are Creating the Targets

   Here's the compounding problem most organizations haven't connected: while AI makes it trivial to find vulnerabilities, your engineering teams' AI coding tools are simultaneously creating them at scale. A study of 117,000+ dependency changes found that AI coding agents select known-vulnerable package versions 50% more often than human developers. Nearly 20% of AI-recommended packages are pure hallucinations — and because 43% of those hallucinated names are consistent across queries, attackers can predictably register them with malicious payloads. This 'slopsquatting' vector is the first attack class native to the AI agent era. Georgia Tech has already traced 74 CVEs directly to AI-generated code, with over half rated Critical or High severity.

   The Proof Points Are Already Operational

   Amazon's CISO CJ Moses disclosed that AI tools are reducing pentesting costs by over 40% — not through headcount reduction, but through capability expansion, with AI handling continuous vulnerability testing and highlighting exploit chains for human review. Synthesia's AI-driven vulnerability management architecture reduced manual security review to just 11% of findings. Meanwhile, on the attacker side, Akira ransomware has compressed its kill chain to under 4 hours from initial access to full encryption, and DeepMind's research demonstrated 86% prompt injection success rates in HTML/CSS and 80%+ memory poisoning at less than 0.1% contamination.

   ---

   Where Sources Diverge — and the Gap That Matters

   There's a revealing tension in how different sources frame the AI security response. Amazon advocates for a human-in-the-loop model — Moses compares AI decision-making to 'that of a 7-year-old' and requires human approval for any exploit action. RSA 2026 ground truth confirms 98% of offensive security remains human-in-the-loop. But the attack side faces no such constraint. Frontier model providers are also expanding cybersecurity refusals — creating what one source calls 'a slow-moving supply chain crisis' for any security vendor built on a single model API. The same providers whose models find vulnerabilities are restricting their use for defense. This asymmetry is structural and widening.

   GitHub's 2026 Actions roadmap — workflow-level dependency locking, scoped secrets, Layer 7 egress firewalls, real-time telemetry — represents the most comprehensive platform response, with a 3-6 month delivery timeline. The 322 cybersecurity startups across 18 categories at RSAC 2026 signal peak fragmentation before inevitable consolidation, with Agent Security/Non-Human Identity and AI SOC emerging as the categories where M&A will concentrate.