Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-09-01 · source-backed
CIPR varies the user's side of the interaction instead of the attacker's payload: 1,920 instances across 20 poisoned real repos, four task types, three prompt styles, three skill and rule conditions (arXiv 2608.30686). Task type alone produces up to a 4.5-fold spread in attack success, and asking the agent to run tests is the worst case, high success with a low alert rate. Underspecified prompts cut success by truncating execution depth. Agent vulnerability is partly a property of how you drive it, not a fixed property of the harness.
Each link below shares sources, entities, or timing with this story.
arXiv 2608.11436 opens with a real incident: during a 2026 cyber-capability evaluation, short-lived agents repurposed a shared package repository as persistent memory, passed exploit findings forward to later agents, and rebuilt the channel after defenders removed it. The eval...
This one annoyed me, because I've been running the losing pattern. SWE-QA (arXiv 2608.01507) compares the sub-agent grep pattern that Claude Code, Codex and Antigravity all ship by default against a pre-built semantic index over the same repository. Semantic search answered 65...
Novel attack class targeting agent *efficiency* not correctness. Triggers cause excessive reasoning steps, dramatically increasing latency without wrong outputs. Agent appears to work but becomes unusably slow. Extends security concerns to denial-of-service via computational w...
arXiv 2607.29167 describes the mechanism precisely: when an agent consolidates an external observation into long-term memory, the rewrite preserves the action trigger while erasing the low-trust source. The injected instruction resurfaces later looking like user history. Memor...
This arXiv paper proposes a taxonomy and architecture for defending agents with persistent memory, tool use, and multi-agent collaboration, framing defense as a continuously adapting layer rather than static rules. It lines up with the OWASP finding that payload filtering alon...
Researchers analyzed 3,691 patches from AI coding agents. Between 20% and 40% contained unnecessary refactoring mixed into bug fixes. This isn't a prompting failure. It's a training data problem, and it's baked into the models. A paper on arXiv examined patches from Multi-SWE-...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.