Latest updates · 21 October 2026
In the full issue →Autonomous AI agents are now breaking into company websites: a Chinese-speaking criminal pointed agents at more than 1,200 sites with almost no human help, while defenders get a new open-weight security model.
Research, industry & policy
- Hackers use autonomous AI agents to breach company websites, The Hacker News · 20 Oct 2026: The agents targeted more than 1,200 sites and broke in with almost no human help.
- OrcaRouter releases OrcaCyber Nano 1.0, OrcaRouter via PR Newswire · 20 Oct 2026: A 27-billion-parameter open-weight cybersecurity model for coding, reasoning and defensive security on consumer-grade hardware.
Journals & preprints
- Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents, arXiv 2610.26438 · 19 Oct 2026: Large language model (LLM) agents now trade real capital using news as evidence, yet they cannot reliably tell fabricated headlines from real news.
- SpanGuard: Decoupling Instruction Localization from Authorization for Lightweight IPI Defense, arXiv 2610.26013 · 19 Oct 2026: LLM-based guardrails are effective against indirect prompt injection (IPI), but their performance depends heavily on model capability: as the guardrail model becomes smaller or weaker, detection degrades substantially.
Conferences, workshops & events
- CSA AI Security Summit, 16 Nov 2026 · San Francisco: Cloud Security Alliance.
Why it matters
Attackers using agents at scale means defenders need automated monitoring and patching that can keep pace.
