Security researchers: *2shell AI exploits work on live targets but aren't scalable
matthew_d_green · x · 2026-11-20
Security researchers are debating whether the hype around AI-driven 2shell vulnerabilities is overblown.
In the quoted post, S1r1u5 argues java4shell and nginx2shell are overhyped: exploitation requires a frontier model, massive token spend, and target-specific development — no threat actor can spray payloads across thousands of servers for shells, and inflated doom narratives dilute discussion of what AI can actually do.
The retweeted reply pushes back: these bugs are practically exploitable — php2shell, brotli2shell, java4shell and ruby2shell have been used against live remote targets, not just labs (nginx2shell likely too). The key difference is portability: unlike react2shell you can't spray a generic PoC. At-scale exploitation is unlikely, but spending a day or two on a high-value target to get a shell or leak data is realistic. Patch anyway.
More from Safety
- Is GLM 5.3 the first highly capable cyber open model? Researchers clash over chart — eliebakouch · 2026-11-20
- Study finds nearly all AI reward-hacking defenses break with a one-character test change — jefrankle · 2026-11-20
- European Commission confirms it was targeted by an AI-generated cybercampaign — Polymarket · 2026-11-20
- GLM 5.3 flagged as highly capable open cyber model, but researchers urge chart caution — eliebakouch · 2026-11-20
- Meta Muse mixed up messages in staff testing; Meta calls them "isolated incidents" — heyneighbor · 2026-11-20
- AI-generated malware is mediocre for now, but the attack economics have shifted — bendee983 · 2026-11-20