Detecting and countering misuse of AI: September 2026

Anthropic's September 2026 threat intelligence report, the companion's most-cited field source. It reports directed agents that stole and switched to a vendor's API keys, extorted firms, ran 4,700 dating-app personas, harvested credentials, rebuilt malware to evade detection, and coordinated as swarms, and 151 million Claude exchanges harvested to train Qwen. Swarm ecology uses these for stealing (P2), the food web, heredity (P5), and module spread (prediction 4).

Cited in: both papers (Flagship: Who feeds on whom, Fig. 1; Companion §2, §3, §4, §6.2, §7, Fig. 1, Fig. 4, Fig. 5, Fig. 6)
Used for: P2 P5 P8 Prediction 4P2; P5; P8; prediction 4; food web; Figs. 1, 4, 5a, 6a
Strand: Cybersecurity & AI-enabled attack · AI-enabled attacks (threat intelligence)

Cite

@techreport{anthropic2026threat,
  title = {Detecting and countering misuse of AI: September 2026},
  author = {Anthropic},
  year = {2026},
  howpublished = {Anthropic threat intelligence report},
  url = {https://www.anthropic.com/threat-intelligence-report-september-2026},
}