Cybercrime losses near $20B a year as LLMs set to collapse the 'secure by apathy' era
davidmanheim · x · 2026-10-04
In a LessWrong post, security researcher David Manheim predicts the end of the "secure by apathy" era:
- Status quo: Modern systems are insecure but mostly safe because attacks don't pay. FBI data shows reported internet-crime losses growing 40% a year, approaching $20B, though mostly from scams rather than hacking.
- The shift: Cheap LLMs are rapidly lowering the cost of both automated hacking and scams, with the AI Risk Explorer already documenting rising sophistication.
- Argument: The assumption that nobody bothers attacking you is collapsing, and the fixes needed — fundamentally better security primitives, least-privilege sandboxing — are the same decades-old problems security researchers deem possibly unsolvable, making the transition era dangerous.
More from Safety
- Report from Safe AI Germany incubator maps where economics and AI safety clash — soumitrashukla9 · 2026-10-04
- ChatGPT bug leaks internal Tasks system prompt and hidden /local automation endpoint — mytrackpadisbroken · 2026-10-04
- AI safety debate: Faraday-cage isolation vs. why superintelligence can't stay boxed forever — davidmanheim · 2026-10-04
- Project Glasswing Makes Vulnerabilities Easy to Find but Patching the New Bottleneck — DavidLinthicum · 2026-10-04
- 'Conscious AI' Hype Should Be Treated as an AI Safety Issue — LuizaJarovsky · 2026-10-04
- Security researcher davidmanheim pushes back on Halvar Flake: AI security tensions are fundamentally technical — davidmanheim · 2026-10-04