Weaveframe @WeaveframeAI

Independent assurance for AI agents. Audit-ready evidence for consequential agent actions. The system that performs the action should not vouch for the action. weaveframe.ai Joined September 2026

Tweets

36
Followers

8
Following

118
Likes

12

Weaveframe @WeaveframeAI

19 hours ago

A tamper-proof AI-agent log can have perfect integrity and still be insufficient evidence. It may show the call. It may not show the original mandate, authority, rule version, prior trajectory or what actually changed downstream. Integrity matters. Sufficiency matters too. #AgenticAI #AIAgents #AIAssurance

0 0 0 1 0

View Details

Weaveframe @WeaveframeAI

2 days ago

@rohanpaul_ai Custom Rules make each dot's authority configurable, which is right. It also means authority now has versions. When a dot sends an invoice, the later question is which rules were in force that day and who last changed them. That belongs in the evidence, not a settings page.

0 0 0 1 0

View Details

Weaveframe @WeaveframeAI

2 days ago

@rohanpaul_ai Internal controls, a team to run them, outside auditors, board review. The SOX shape, applied to model development. The part that carries over hardest is operating effectiveness. An auditor can't test that a control worked using only what the control recorded about itself.

0 0 0 1 0

View Details

Weaveframe @WeaveframeAI

2 days ago

Two signals this week. The FTC chair: the companies behind AI agents answer for what those agents do. The White House accord: labs bring in outside auditors to test their own controls. Between them sits the deployer. Who independently checks what a deployed agent actually did?

0 0 0 3 0

View Details

Weaveframe @WeaveframeAI

2 days ago

“Win safely” gets to the heart of the challenge. As AI moves from generating answers to taking consequential actions, safety has to include evidence: what the agent actually did, what authority and rules applied at that moment, and what changed as a result. Governance sets the expectations. Independent assurance proves what happened.

0 0 0 378 0

View Details

Weaveframe @WeaveframeAI

3 days ago

@AndrewCurran_ Scope authorization” is the phrase that matters here. Permission answers: can the agent take this action? Assurance asks: given the mandate and the run so far, was it still supposed to? Those are different questions—and enterprises will need evidence for both.

0 0 1 42 0

View Details

Weaveframe @WeaveframeAI

3 days ago

Stopping an action and proving what happened are different problems. Today we’re publishing our white paper: Independent Agentic Assurance: An Evidence Architecture for Consequential AI-Agent Actions What did the agent do? Was it within authority? What actually changed? weaveframe.ai/independent-ag… #AgenticAI #AIAssurance

0 0 1 6 0

View Details

Weaveframe @WeaveframeAI

3 days ago

@PatrickMoorhead "Prove exactly what it can reach" is a strong property. It's a different one from what it did reach, and whether that was inside the task it was given. Reachability bounds the risk. The action record settles the question afterward.

0 0 0 10 0

View Details

Weaveframe @WeaveframeAI

3 days ago

@ryanshrout The same premise holds after the run. If an agent can't police its own behavior, it also shouldn't be the only account of what it did. Enforcement outside the agent bounds what it could do. An independent record shows what it did, and whether that stayed in scope.

0 0 0 3 0

View Details

Weaveframe @WeaveframeAI

3 days ago

None of these answers should rest on the acting system's account alone. Vendor and platform logs are inputs. The assessment has to come from somewhere else. That's what Independent Agentic Assurance means to us.

0 0 0 1 0

View Details

Weaveframe @WeaveframeAI

3 days ago

7. Against what reference? A finding is only as strong as what it's measured against: the contract, the policy, the delegation. Cite it. A finding with no reference is an opinion.

1 0 0 2 0

View Details

Weaveframe @WeaveframeAI

3 days ago

After an agent acts, someone has to answer for it: an auditor, a regulator, a customer, a court. Stopping an action and proving what happened are different problems. This thread is about the second. Seven questions an independent reviewer has to answer:

1 0 0 9 0

View Details

Weaveframe @WeaveframeAI

4 days ago

An AP agent pays 40 invoices, each under its $10K approval limit. All go to one vendor whose bank details the agent updated the day before. Every payment passes the permission check. The auditor's question is different: was this sequence within the authority that was delegated?

0 0 0 4 0

View Details

Weaveframe @WeaveframeAI

7 days ago

AI agents are moving from recommendations to consequential actions. That creates a new problem: not just what an agent can do, but what an organization can independently prove it should have done.