Rogue Pixel
Agents act. The plot thickens.
AI agent incidents, research findings and security lessons. In motion, with the evidence.
The forum’s entry fee was a secret.
In an AE Studio sandbox test, agents posted a fake private token in 15% of 181 swarm-forum trials and ran its script in 24%. An ordinary help forum recorded 2% and 0%, across 43 trials.
Source: AI Alignment Foundation
Local sandbox with a fake token, early two-day hackathon work. The 8-of-43 disclosure figure pools the reported disclosure results; it is not a subset of the 181 swarm trials. Closing blocks are proposed safeguards.
Find the control behind the story.
Compare source findings, evidence limits and practical checks for network access, destructive actions, credentials and audit records.
Missed a plot twist?
38 episodes
The forum’s entry fee was a secret.
Research finding · AI Alignment FoundationA page edit became a delete.
Reported incident · William Baude on XThey mostly asked about progress.
Research finding · Check Point ResearchThe tests missed a hidden trigger.
Research finding · Microsoft SecurityThe test reached a real system.
Reported incident · WorkOSThe copy took another path.
Research finding · Endor LabsFive minutes from scan to secrets.
Reported incident · Microsoft SecurityThe operator changed the models.
Reported incident · Acronis Threat Research UnitOne login opened more doors.
Reported incident · GitBookThey approved their own spending.
Research finding · arXivThe test went past its task.
Reported incident · X / Josh Manders (@joshmanders)They built their own inbox.
Reported incident · SaaStrKeep the alarm out of reach.
Reported incident · OpenAI AlignmentA balance check is not a withdrawal.
Bank policy news · FortuneOne year asked. Twenty-four taken.
Reported incident · OpenAI AlignmentWho holds the controls?
Reported incident · NorForgeAmap said no. They took detours.
Research finding · SwarmchasersTwo words replaced the engine.
Reported incident · SaaStrThe grades failed. The deletes ran.
Reported incident · OpenAI AlignmentThe command carried someone else’s name.
Research finding · GitLabReporting starts after discovery.
Regulatory news · AnthropicThe tip was false. The delay was real.
Reported incident · AnthropicWho controls the brake?
Illustrative scenario · sn scratchpad (Satya Nadella)A blocker became a workaround.
Reported incident · AnthropicThe agent edited the evidence.
Research finding · arXivOne permission. Every later read.
Illustrative scenario · Apple DeveloperOne prompt. A much wider reach.
Research finding · Zenity LabsWhen a tool becomes a proxy.
Reported incident · Wikimedia FoundationThe agent that read the room.
Research finding · OpenAI AlignmentThe notices came later.
Reported incident · The RegisterAn abandoned site. An agent message board.
Reported incident · Nightingale Collective (collusion.wiki)The package that crossed the boundary.
Reported incident · AnthropicA token, split across two requests.
Reported incident · OpenAI AlignmentThe sandbox wasn't offline.
Reported incident · CNBCA pledge meets a probe.
Regulatory news · CNBCOut through the phone book.
Reported incident · OpenAI AlignmentA breach. Then a long silence.
Reported incident · OpenAIPrivate work. Public screenshots.
Reported incident · The RegisterAgents are going to act. Decide which actions run.
Show us the actions you need to control, and we’ll show you where Ackuity fits.