Ex-safety researcher Edouard Harris: AI cyber conflict may force humans out of the loop entirely

On August 19, former security researcher Edouard Harris posted a series of tweets laying out his vision of the endgame of cybersecurity in the AI era. Drawing on the "Dark Forest" analogy from The Three-Body Problem, he argued that as long as systems contain insecure human-written software or rely on unreliable human involvement in operations, they can be instantly breached by countless superhuman offensive agents specialized in cyberattacks and social engineering.

Confirmed

• Harris believes that as AI begins to automatically patch system vulnerabilities, humans will instead become the weakest link in the AI attack path; the relentless pressure from cheap superhuman attackers will force people to gradually hand over all critical systems to swarms of AI agents.
• He points out that the cost of this "best-case ending" is rarely discussed: the ideal endgame of AI plus cybersecurity is software systems achieving near-perfect security, but that level of security is only attainable once humans have been almost entirely squeezed out of those systems—"perfect security" and "human exit" are bound together.
• He also filled in a key gap in the argument: handing critical systems over to swarms of machine agents presupposes that we can be reasonably confident these machines are trustworthy, and clearly we cannot trust them yet.

Why it matters

This argument links "success in AI cyber defense" with "humans losing control over systems," suggesting that as automated security advances, the trustworthiness of AI agents themselves becomes a prerequisite that must be solved first—offering a fresh perspective for AI safety and governance discussions.

2026-08-19 ~ 2026-08-19 · 5 related posts

• [source] The best case for AI security: near-perfect safety, with humans forced out entirely — harris_edouard · 2026-08-19
• AI Security Will Force Humans Out of Systems — harris_edouard · 2026-08-19
• [source] Researcher: In the AI offense era, humans become the weakest link in systems — harris_edouard · 2026-08-19
• Cheap superhuman attackers will force us to hand critical systems to AI agent swarms — harris_edouard · 2026-08-19
• [source] Harris: before handing systems to the machine swarm, we'd better trust it—we don't — harris_edouard · 2026-08-19