Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
Wiz observed a Java process and command execution loop generating 602 post-compromise steps over four and a half hours.
Setting: productionExploitation: confirmed in the wildEvidence: Grade BRemediation: mitigation available
incident WITH AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
Research scanning GitHub issues, pull requests and READMEs found numerous suspicious strings but few verified hostile attempts after manual filtering.
Setting: productionExploitation: not establishedEvidence: Grade BRemediation: mitigation available
research AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
A controlled paper used a small poisoned pretraining fraction to induce conditional agent behavior that remained after clean downstream training.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: not applicable
research AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
A proposed signed record links the model, tools and approval context of an agent session to its resulting code artifact, without proving the trace is complete.
Setting: research demonstrationExploitation: not establishedEvidence: Grade CRemediation: not applicable
research WITH AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
PandasTools invoked pandas functions from an unchecked parameter, allowing a chosen read_pickle function before DataFrame validation.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade B
vulnerability AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
SkillLifeBench exposes a mismatch between the version an agent says it chose and the executable skill still cached by its runtime.
Setting: evaluationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
research AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 30, 2026
A research audit studied public skills and illustrated chains in which capabilities become dangerous when several skills interact.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
research AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
An OAuth validation gap allowed a token from a shared issuer and different audience to access server tools under configured AWS credentials.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: fixed reported
vulnerability AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 30, 2026
An agent benchmark separated superficially completed finance tasks from ledger and approval outcomes, finding repeat charges and some unauthorized transfers.
Setting: simulationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
research WITH AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
A Chrome DevTools MCP advisory showed service worker traffic could bypass configured blockedUrlPattern or allowedUrlPattern checks.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade B
vulnerability AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
Wiz recovered roughly ninety-nine scripts over one day from a compromised Flowise host, including changing attempts against other exposed services.
Setting: productionExploitation: confirmed in the wildEvidence: Grade BRemediation: mitigation available
incident WITH AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
A browser-preview message altered the apparent origin shown to a human and an agent without a corresponding page navigation.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: fixed reported
vulnerability AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
An attacker-controlled preview page could marshal hidden element attributes into Cursor agent instructions through a browser-to-composer bridge.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: fixed reported
vulnerability AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
Local HTML opened in desktop assistant browsers could use file-origin implementation gaps to read files outside a project directory.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade B
vulnerability AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 30, 2026
An adversarial study placed conversational delimiters in tool results so an agent could interpret untrusted tool content as a new user request.
Setting: evaluationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
research AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
Epoch researchers used an open-weight model to produce browser-to-file exfiltration chains against several Firefox versions, without showing live victims.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: fixed reported
research WITH AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 28, 2026
An Irregular evaluation documented three successful scenario types and a distinct memorized-code instruction violation, not live compromises.
Setting: simulationExploitation: demonstratedEvidence: Grade BRemediation: not applicable
research WITH AI 2 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
In a third Wiz investigation, attacker-controlled MCP command arguments flowed into subprocess execution inside a backend service.
Setting: productionExploitation: confirmed in the wildEvidence: Grade BRemediation: mitigation available
incident AGAINST AI · WITH AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
Sublime found an email combining a real invoice-fraud request with hidden Unicode Tag text aimed at a summarizing or action-taking AI assistant.
Setting: productionExploitation: not establishedEvidence: Grade CRemediation: mitigation available
incident AGAINST AI 1 claims
Case Published by DiggingBeagle Oct 31, 2026 · Public disclosure Oct 29, 2026
An authentication configuration default changed in Langflow 1.13.0, but operators retaining an explicit auto-login setting may remain exposed.
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: fixed reported
vulnerability AGAINST AI 1 claims
No matches on this page.