Index · updated 2026-10-13
AI Security Index
The Injection has tracked 57 AI security stories from 31 sources since August 27, 2026. 26% were rated high or critical, 14 came with a CVE (median CVSS 7.7), and 119 researchers are credited by name.
Every figure is computed from the verified feed on each build. Severity is The Injection's editorial rating (method); weeks follow disclosure dates. Each section has a citation you can copy.
57
Stories
15
High or critical
14
With a CVE
31
Sources
119
Researchers credited
10
Products tracked
How severe?
26% of the AI security stories The Injection tracked were rated high or critical (15 of 57).
- CRIT1
- HIGH14
- MED15
- LOW2
- INFO25
What kind?
Vulnerability is the largest category, with 17 of 57 stories (30%).
Which products?
Claude apps appears in more AI security stories than any other product on The Injection: 9, ahead of Claude Code (7).
Which techniques?
AI agent security is the most common technique, in 14 of 57 stories (25%).
How fast?
49 AI security stories were disclosed in the last 4 weeks, up 513% from the 4 weeks before (8).
CVEs
14 AI security stories on The Injection came with a CVE; the median CVSS score is 7.7.
Top sources
- 1Anthropic6
- 2AIR Security5
- 3arXiv4
- 4Socket3
- 5UK AI Security Institute3
- 6Zenity Labs3
- 7CrowdStrike2
- 8Google Threat Intelligence Group2
- 9JFrog2
- 10Microsoft2
Most credited
- 1Anthropic3
- 2Avishai Efrat2
- 3Dor Granat2
- 4Eliad Mualem2
- 5João Donato2
- 6Or Nevo2
- 7Socket Research Team2
- 8Xander Davies2
- 9Abby D'Cruz1
- 10AI Security Institute1