Anthropic on Cyber: Open Capable Models Haven't Sparked a Hacking Explosion Yet
joshua_saxe · x · 2026-10-17
Anthropic's Logan Graham explains why cyber-capable open models haven't yet caused a hacking explosion:
- Offense and defense co-proliferate: capabilities spread to attackers and defenders alike, and defenders may be keeping up.
- Signs of vertical growth: some indicators went vertical from March to August this year (at least prior to GLM 5.3), so it's premature to rule out a meaningful rise.
- Case in point: a relatively amateur attacker in South Korea used existing frameworks to breach an institution, alongside recent bank attacks in South Korea and Japan.
Rohit Krishnan adds: for AI-savvy people, doing AI for money is far more lucrative than cybercrime — an economic brake on malicious pivots.
More from AGI Musings
- AI filmmaker claps back: "without AI, my films simply wouldn't exist" — PurzBeats · 2026-10-17
- Murray Shanahan: LLM whisperers' strange findings still ignored by academia and labs — mpshanahan · 2026-10-17
- Ramp's $60B Valuation Bets AI Agents Will Run Corporate Finance — LexSokolin · 2026-10-17
- Bittensor expands subnet capacity to 2,500 miners, opening autonomous AI agent participation — bittingthembits · 2026-10-17
- Deepfakes Are Everywhere, Yet Their Real Impact on Elections Fell Far Short of Fears — binarybits · 2026-10-17
- AI was supposed to make software cheaper. Incumbents used it to raise prices — then agentic coding evaporated the moat in 5 months — akumar · 2026-10-17