AI sovereignty is incomplete without action sovereignty.
Owning models, data, and compute matters.
Reading paths
Choose a reader path, then narrow by series, topic or keyword. Articles are ranked by editorial priority, role fit, topic match, recency and diversity.
Want everything in date order instead? Browse all news.
Reader path
{{ featuredReason }}
{{ featuredMatch }} match
Owning models, data, and compute matters.
A founder reflection on manufacturing ontology, operational control, and why enterprises should be careful not to surrender their own reality to the AI market.
Final authority, canonical action identity, bounded authorization, and exposure evidence: the current control path and the limits of the proposed BAF/AREG connection.
OSuite Runtime Boundary Benchmark tests whether an agent action is still inside the boundary that was approved, across 20 runtime surfaces, 38 risk families, 6 locales, and 10 obfuscation styles.
The practical path is not to start with a giant AI policy.
OSuite and Baby Blue / invinoveritas ran a v11 registry-scoped reference run: OSuite created a CAVA action artifact, received a signed independent-mediator verdict, self-submitted it to Baby Blue's ledger, executed a real GitHub issue creation, and checked that the outcome still matched the approved action.
As coding agents move toward longer-running autonomous work, the enterprise control point shifts from whether a model may call a tool to whether the runtime action stays inside an approved boundary.
OSuite and Baby Blue / invinoveritas ran a concrete external-verifier composition: a CAVA action artifact, a partner-signed verdict, and a proof packet that can be checked without trusting either party's internal log.
The paper argues that frontier providers can shape model-side safety, but enterprises still need final authority over the actions taken inside their own environments.
AI governance should not stop at policy language or model posture.
Enterprise AI governance needs to move from policy language and dashboards to enforceable runtime controls for systems that can actually act.
The buyer question is shifting from whether an AI agent is useful to whether the organization can prove what the agent was allowed to do.
A human approval means nothing if a wrapper can rewrite the request after the click.
Microsoft Agent 365 is a useful signal for where the market is going.
Enterprise AI policies usually die in documents.
CAVA gives OSuite a stable action language.
Enterprises do not need another AI demo that disappears when budgets tighten.
Agents are most useful in marketplace operations when they can draft and investigate freely, but need approval before touching users, money, public messages, or policy.
MCP, A2A, ACP, ANP, and trust registries help agents connect.
The point of CAVA is to analyze consequence, not just the raw command.
Why we're building an open protocol for governing autonomous actions, and why OSuite came second.
What remains valid when an agent's environment changes?
A compromised service credential should not become permission to move money.
Two independent agent-security preprints cite our work on action identity and approval binding.
A public replication survey initially marked CAVA's benchmark claim as not released.
Exact-action grants contained the observed scope violations that destination filtering missed.
AI agent harnesses help models plan, call tools, and run longer workflows.
Recording what happened is useful.
Inspired by Suica and FeliCa, we tested what happens when exact-action approval is prepared before dispatch.
An Excel sheet can track agent activity.
If an AI agent can reuse approval without a boundary, the approval becomes a standing permission.
A focused early-access program to bring one governed path, review before launch and replay-ready proof after, to every runtime an enterprise agent touches.
A dashboard shows metrics.
{{ archiveSummary }}
Article series
Follow the argument from buyer objection to architecture, field rollout, strategy and company notes. All series
Topic
Try a broader topic such as runtime, governance, approval, marketplace, or CISO.