Swarm Ecology
layer 3 · theory

web page · AI safety, alignment & autonomy evaluations

How fast is autonomous AI cyber capability advancing?

UK AI Security Institute · 2026

AISI blog

A UK AI Security Institute post asking how fast autonomous AI cyber capability is advancing. The companion cites its finding that the length of cyber tasks that models complete with 80% reliability has doubled roughly every five months since late 2024. It is one of the rising capacities that the argument needs to continue until an agent can pay its own way.

Cited in
the companion paper (Companion §1)
Used for
Capability trends (§1)
Strand
AI safety, alignment & autonomy evaluations · Autonomy, replication & dangerous-capability evaluations

Cite

@misc{aisi2026howfast,
  title = {How fast is autonomous AI cyber capability advancing?},
  author = {UK AI Security Institute},
  year = {2026},
  howpublished = {AISI blog},
  url = {https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing},
}