Anthropic researcher predicts AI can harden world's cyber risk in 1-2 years; security engineers push back
joshua_saxe · x · 2026-10-18
Anthropic researcher Logan Graham predicts that within 1-2 years AI tools could "harden the world," finding all bugs, writing secure-by-default code, and defending everything.
Security engineer S1r1u5 disagrees: writing more code is generally bad — systems should be designed by minimizing security-critical code and building strong isolation boundaries. Using V8 as an example, rewriting an entire engine to eliminate all JIT bugs seems far beyond current model capabilities, and models obsolete in months leave behind code with the same vulnerability problem.
The thread captures a real split between AI-driven security hardening optimism and system-design conservatism.
More from Models
- DeepSeek Ships V4.1 Flash 1017: Gains on All 11 Benchmarks at 1/16 the Price of Rivals — Prompt Engineering · 2026-10-18
- Redditor claims Gemini 4 Argon made a 700-page book — screenshot shows it pushed back — opmgyhx · 2026-10-18
- Same prompts cost ~3.3k tokens on AI Studio but drain 16% of Gemini App's 5-hour limit — PECman1728 · 2026-10-18
- Gemini 4 Argon takes top spot on Edison Sci's long-horizon bioinformatics benchmark BixBench3 — divy93t · 2026-10-18
- Unsloth releases Dynamic v3 GGUF quants for LiquidAI's LFM2.5-24B-A2B — helloiamleonie · 2026-10-18
- DeepSeek 4.1 Flash wows with procedural Boeing 747 Blender scene renders — victormustar · 2026-10-18