At Black Hat, OpenAI revealed more about the OpenAI / Hugging Face AI agent breach. One of their key takeaways for defenders:

“We should invest as well additionally in slowing down offensive agents. Things like honey tokens and deception can introduce uncertainty into the agents’ behavior. Is this credential I found one that I can actually use, or will it trigger a security alert and cause me to be investigated? ... Uncertainty ultimately slows down an attacker.”

It’s a striking recommendation, and it’s aligned with our own research into how offensive agentic behavior changes in the face of deception as well as the guidance published in the immediate aftermath of the breach by the Cloud Security Alliance.

In Hugging Face’s write-up, they note the difficulties imposed by the volume of activities an agent can perform:

We were not dealing with one clever exploit or a clean sequence of attacker actions. They had to correlate thousands of low-signal events across several systems while the agent continued testing new paths. The successful path was hidden inside the noise generated by the thousands of failed ones.

The value that deception provides – in the form of high-fidelity alerts that cut straight through the noise and reveal precisely which system has been compromised – feels more critical than ever in the face of this entirely new class of attack.

We agree with the conclusion of the OpenAI talk:

Threat actors will soon intentionally optimize and deploy agent collectives.

When attacks are not the unintended result of frontier AI evaluations, but are maliciously targeted by threat actors worldwide without any need for sandbox escape, the scale and urgency of the problem is clear.

How can Tracebit help?

The fact that the teams from Hugging Face and OpenAI have shared so much about this breach gives a unique opportunity for insight into what agent-orchestrated attacks could look like in the real world. In this light, we’ll compare the deceptive capabilities offered by Tracebit to some of the Tactics, Techniques, and Procedures (TTPs) publicly disclosed to see how they might hold up.

Read the full post on tracebit.com →

In June we published research that pointed frontier AI models at a live AWS environment to measure how autonomous AI attackers behave around canaries. The answer: canaries catch AI attackers early and reliably. That was useful, but we wanted to know if we could do more… what if we could disrupt the attacks as they were happening?

This week we published follow-up research with a stronger result. A short string - a context bomb - planted in a canary’s value can do something no alert can: stop the AI agent mid-attack. The strongest attacker, Claude Opus 4.8, went from full admin compromise in 93% of runs to 0%. The model’s own safety guardrails become your defense.

Today we’re shipping both research findings as product. Context Bomb Canaries are now available for AWS environments, and the new AI Posture dashboard gives you one place to manage how AI agents, yours and an attacker’s, interact with your canaries.

Read the full post on tracebit.com →

Frontier AI agents can seize full admin of a cloud account in minutes - a critical action within 14 minutes on average, leaving defenders barely an 8-minute window to respond. When attacks move that fast, detection alone isn’t enough.

So we built something that doesn’t just alert you - it stops the attacker. A context bomb is a short string planted in a decoy secret that trips an AI model’s own safety guardrails, causing the agent to refuse to continue. You still get the alert; the attack halts early.

Across five frontier models and 152 attack runs, a single context bomb cut complete account compromise from 36% of runs to 1%. Opus 4.8, the most capable agent we tested, went from gaining admin in 93% of runs to failing every time. The full benchmark, the data, and what to weigh before deploying one are on the Tracebit blog.

Read the full post on tracebit.com →

Episode 5 of Canaries in the Wild is live - and our first in-person recording. We sat down in central London with Ollie Whitehouse, CTO of the UK’s National Cyber Security Centre (NCSC), the cyber security mission of GCHQ.

Ollie has been running cyber deception for over 23 years, from personal honeypots in 2002 to leading the NCSC’s national-scale deception programme today. He brings a rare perspective that spans hobbyist, vendor, practitioner, and national policymaker.

Read more about the podcast on tracebit.com →

Watch on YouTube, or listen on Apple Podcasts or Spotify

.