Technical disclosure and independent reproduction will determine whether kinetic prompt injection can reliably compromise agents controlling physical systems and trigger safety-relevant physical actions.
state: expiredheat: lowuncertainty: highconvergesscott: mediumagentic-security prompt-injection physical-systemsBlack Hat USA
What is this?
The supplied titles describe a Black Hat USA briefing on “kinetic prompt injection,” reportedly including a live jailbreak of a stock Unitree Go2 robot dog running Gemini to demonstrate that an agent compromise can produce physical consequences. Beyond the event host and demonstration claim, no snippets establish the speakers, attack mechanics, resulting actions, safeguards bypassed, technical disclosure, or independent reproduction, so reliability and safety impact remain unverified.
Why it matters to Scott
The claimed robot jailbreak converges with Scott’s separation-of-powers and runtime-gating position: instruction-bearing sensor input must not carry authority to trigger consequential physical actions. It creates a potential dated-receipts opportunity for his embodied-agent architecture, but the supplied evidence is only a demonstration description; without disclosed mechanics or independent reproduction, it does not yet establish the vulnerability’s reliability or defeat of structural controls.
ip:framework.separation-of-powers-for-cognitionip:framework.decision-authority-infrastructureip:concept.architectural-containmentip:concept.evidence-class-ladderdev:concept.padded-cell-agent-architectureradar:concurrent-audio-prompt-injectionradar:concept.prompt-injectionradar:concept.embodied-airadar:concept.agent-safety
queries asked of Scott's wikis
- prompt injection defenses for embodied agents
- agent permissions and physical action safety
- untrusted sensor input in agent systems
- sandboxing and human approval for robotics agents
- independent reproduction standards for AI security claims
- blast-radius containment in agent harnesses
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-08-11T04:28:51Z
No technical disclosure, attack mechanics, or independent reproduction emerged within the active horizon, so the episode has faded without validating the reliability claim. The architectural warning remains relevant, but this case no longer merits ongoing attention absent substantive new evidence.
2026-08-09T04:23:09Z
The refreshed discussion adds only generic concern and off-topic chatter; it provides no attack mechanics, structural-control bypass, or independent reproduction, so the case’s meaning is unchanged.
2026-08-08T00:30:22Z
No technical disclosure, independent reproduction, or attack-detail evidence has arrived; the small engagement change is merely repetitive attention and does not strengthen the reliability claim.
2026-08-08T00:26:32Z
grounded: converges/medium — The claimed robot jailbreak converges with Scott’s separation-of-powers and runtime-gating position: instruction-bearing sensor input must not carry authority t
2026-08-08T00:24:21Z
origin walked (codex/luna, conf 0.98): anchor reddit.post.1vigldz -> echo.youtube.70b1e3889a by Black Hat
2026-08-08T00:22:54Z
case created — The Black Hat briefing raises a bounded, consequential agent-security claim, but the available observation contains too little technical evidence for higher confidence or heat.
Decision trace
- 08-11 14:28expireNo technical disclosure, attack mechanics, or independent reproduction emerged within the active horizon, so the episode has faded without validating the reliability claim. The architectural warning r
- 08-11 14:28alert_silentThe staleness trigger adds no consequential fact; the original demonstration was already routed, and renewed attention should wait for disclosed mechanics or independent reproduction.
- 08-11 14:28alert_routeThe staleness trigger adds no consequential fact; the original demonstration was already routed, and renewed attention should wait for disclosed mechanics or independent reproduction.
- 08-09 14:23repriceThe refreshed discussion adds only generic concern and off-topic chatter; it provides no attack mechanics, structural-control bypass, or independent reproduction, so the case’s meaning is unchanged.
- 08-09 14:23alert_silentThe public demonstration was already routed, and the new delta is merely low-information discussion rather than a consequential technical fact.
- 08-09 14:23alert_routeThe public demonstration was already routed, and the new delta is merely low-information discussion rather than a consequential technical fact.
- 08-08 10:30repriceNo technical disclosure, independent reproduction, or attack-detail evidence has arrived; the small engagement change is merely repetitive attention and does not strengthen the reliability claim.
- 08-08 10:30alert_silentThe concrete Black Hat demonstration was already routed, and this look adds only negligible engagement rather than a consequential new fact.
- 08-08 10:30alert_routeThe concrete Black Hat demonstration was already routed, and this look adds only negligible engagement rather than a consequential new fact.
- 08-08 10:29alert_shadowThe official recording establishes that a concrete physical-agent compromise demonstration was publicly presented, directly supporting immediate review of sensor-input trust boundaries and runtime act
- 08-08 10:29alert_routeThe official recording establishes that a concrete physical-agent compromise demonstration was publicly presented, directly supporting immediate review of sensor-input trust boundaries and runtime act
- 08-08 10:26groundThe claimed robot jailbreak converges with Scott’s separation-of-powers and runtime-gating position: instruction-bearing sensor input must not carry authority to trigger consequential physical actions
- 08-08 10:24promote_anchororigin walk conf 0.98
- 08-08 10:22createThe Black Hat briefing raises a bounded, consequential agent-security claim, but the available observation contains too little technical evidence for higher confidence or heat.