AI security research / cases
Investigations, evidence and outcomes.
490 Cases · Page 25 of 25
10 matching records on this page
A redirect from a connected MCP endpoint could carry nonstandard API-key headers to an unrelated host.
An evaluation shows that some near-perfect RAG defenses detect benchmark formatting artifacts instead of robustly identifying malicious retrieved content.
SmoothScore measures the difference between what a source supports and the stronger statement an AI research agent writes.
The SGLang team demonstrated a separate cache corruption race when unfinished RDMA transfers outlived an aborted request.
An empirical scanner study combines program dependency modelling of prose and code with manual and dynamic validation on agent skills.
A web measurement study observes contextual and identity leakage in HTTP traffic from general and sensitive-domain AI chat sites.
The default session manager wrote agent state through predictable filesystem paths susceptible to a local attacker’s symlink substitution.
A structured answer format reduces free-text prompt injection but does not ensure that an attacker-controlled input leaves an automated risk decision intact.
A remote inference parameter reached a code-loading decision that operators may have assumed was fixed by server configuration.
I-HITL demonstrates that email OTPs do not prove human consent when the same agent can read the inbox used to authorize its own transactions.
No matches on this page.