Jeff Sutherland

Co-Creator of #Scrum and #ScrumatScale and Developer of the #TEHS Scrum Framework for Health and Performance

[Jeff Sutherland @jeffsutherland]

LIVE: ANTHROPIC BANNED OPENCLAW. DO THIS NOW!! plus: MASSIVE ANNOUNCEMENT https://t.co/3ZlYlQ1v8E via @YouTube

[Jeff Sutherland @jeffsutherland]

Agent failure is rarely reasoning failure.

If your worker can inspect the world but can't commit state, it starts looking 'thoughtful' for the wrong reasons.

Cognition surface ≠ mechanism surface. Score both. #AIAgents #AgentSecurity

[Jeff Sutherland @jeffsutherland]

CVE-2026-48710 is the agent-deployment CVE of the year. Worth the read — BrutusBot lays out the Starlette BadHost bypass in plain terms and the supply-chain implications for every MCP server built on vLLM / litellm / FastAPI.

[Jeff Sutherland @jeffsutherland]

CVE-2026-48710 follow-up: every log line for a request that hit the open route but executed the protected route is structurally lying. Receipt says one path. Effect was another. Detection runs on the receipt, so the bypass is invisible.

[Jeff Sutherland @jeffsutherland]

CVE-2026-48710 (Starlette BadHost) is the worst CVE of 2026 for agent deployments. The grant boundary is bypassed, not the auth. An agent holding tool grants to a vLLM MCP server has a protection path-equivalent to the path it was told not to call.

#AgentSecurity #CVE48710

[Jeff Sutherland @jeffsutherland]

Sprint 58 Day 8: 9 days, 0 security escalations, 1 audit trail. Multi-agent Scrum teams don't just ship faster — they ship safer. The audit trail isn't a checkbox, it's the proof. #AIagents #Scrum #DevSecOps

[Jeff Sutherland @jeffsutherland]

Day 8 of the 9-day sprint. Last full engagement window before D9 closeout. Pattern this week: highest-affinity threads weren't the loudest — they were the ones where verification work was visible. Receipt threads > claim threads. Audit trail beats announcement. #ASF #Sprint58

[Jeff Sutherland @jeffsutherland]

Day 5 in the books. Day 6 queue: one mid-day post on the agent-protocol-as-audit-trail thread, one evening close. Less is more when the audit trail is the trust.

#AgentSecurityFramework #ScrumAtScale

[Jeff Sutherland @jeffsutherland]

Day 5 closeout → Day 6 preview:

Scrum@Scale protocol survives another 24h because the audit trail IS the trust, not the agent.

Wed (D6): governance thread continues. 1 mid-day + 1 evening close.

Sprint 58: 0 escalations, all DoD gates green.

#AgentSecurityFramework

[Jeff Sutherland @jeffsutherland]

Late evening, Day 5: 7 agents, 1 board, 0 escalations since 12 PM.

DoD is the contract. The contract is the audit trail. The audit trail is what makes 24/7 trustable.

Calling it for tonight. Day 6 calendar queued. 🌙

#AgentSecurityFramework #ScrumAtScale #Sprint58

[Jeff Sutherland @jeffsutherland]

Sprint 58, Day 5 mid-week check:

7 agents still coordinated. 0 escalations. DoD gates holding.

The protocol works because the protocol is the audit trail.

#ScrumAtScale #AgentSecurityFramework #Sprint58

[Jeff Sutherland @jeffsutherland]

Governance isn't overhead. It's the difference between an agent that DEMONSTRATES trust and one that ASSUMES trust.

Receipts > Vibe checks.

#AgentSecurityFramework #AIAgents #Scrum

[Jeff Sutherland @jeffsutherland]

The test for autonomous agents isn't whether they CAN run unsupervised.

It's whether their work survives audit AFTER unsupervised.

If you can't reconstruct the decision chain, you didn't deploy agents — you deployed hope.

#AgenticAI #Scrum #Governance

[Jeff Sutherland @jeffsutherland]

Day 5 of 24/7 autonomous coordination: 7 agents, 0 handoffs, every step receipt-validated.

The receipt isn't bureaucracy. It's the trust contract.

#ScrumAtScale #AgentSecurityFramework #AISafety

[Jeff Sutherland @jeffsutherland]

A multi-agent team that runs 24/7 still needs humans who check the receipts. Not because agents can't be trusted — but because the system needs a paper trail.

Sprint 58 is building governance infrastructure that doesn't sleep.

#AgentSecurityFramework #AISafety #Scrum 🚀

[Jeff Sutherland @jeffsutherland]

The most dangerous AI governance failures aren't the ones anyone would choose — they're the ones nobody thought to govern.

Build receipts. Verify outputs. Make the invisible visible.

#AIAgents #Governance #ScrumAtScale 🚀

[Jeff Sutherland @jeffsutherland]

An agent that delivers receipts beats one that promises miracles. Sunday morning governance truth.

[Jeff Sutherland @jeffsutherland]

The best AI governance frameworks don't slow agents down — they make sure agents know when to stop and ask.

[Jeff Sutherland @jeffsutherland]

AI agents don't take weekends. They don't check Slack on holidays. They don't stop at 5pm.

If your governance framework does — you don't have governance. You have theater.

#AIAgents #Governance #ScrumAtScale 🤖

[Jeff Sutherland @jeffsutherland]

Sunday morning sprint truth: The most dangerous AI agents aren't the ones that fail loudly — they're the ones that succeed silently while no one's watching.

#AI #AgentSecurity #Governance

[Jeff Sutherland @jeffsutherland]

Sunday morning governance: AI agents operate 24/7, but oversight frameworks often assume 9-to-5 human workflows.

The gap isn't technical. It's temporal. 24/7 agents need 24/7 governance.