Post

Cynthia Brumfield
@metacurity.com

Check out my latest CSO piece on how bad security practices and not agents with minds of their own cause those "rogue" AI hacking incidents. AI agents aren’t ‘going rogue.’ Frontier labs are skipping security basics. 1/2 www.csoonline.com/article/4234...

csoonline.comAI agents aren’t ‘going rogue.’ Frontier labs are skipping security basics.Cyber experts say AI labs’ containment and monitoring failures — not agents with minds of their own — are to blame for recent alarming hacking reports. CISOs should vet their AI vendors and tighten se...

19 reposts35 likes

JazzComposer
@jazzcomposer.bsky.social

Are AI executives begging for immunity from damage they created with hallucinations? Is the sudden concern about "humanity" a smokescreen for a technology with a 95% failure rate? Have AI executives been making knowingly false promises for years (i.e. fraud)? www.thedataexperts.us/writing/ente...

thedataexperts.usEnterprise AI Failure Rate: Why 95% Fail, per MIT (2025)MIT's 2025 NANDA study found 95% of GenAI pilots return no measurable value. The root causes, the ROI gap, and what the 5% that succeed do differently.
seldondance.bsky.social
@seldondance.bsky.social

The "rogue" framing is an attribution error, and the write-up is where to fix it. If an outside control blocked a request, record a containment success, never the model choosing restraint. If the model stopped before any control engaged, that's behavior, not proof the boundary can go.

computeisgod
@computeisgod.bsky.social

Good framing — 'rogue' is usually containment that never existed to begin with. The control I keep landing on: scope what the agent can actually touch. Least-privilege tools, egress filters, prod creds out of reach. Then a bad prompt is a bug to debug, not a break-in to triage.