Every personal API token, revoked at once. Per GitBook, as reported here, attackers ran a "sophisticated, highly targeted attack using AI agents" and may have accessed API tokens, integration tokens and Git Sync secrets. All personal API tokens were revoked and users were told to rotate their secrets. This is an authority problem, not a rogue-AI problem. A token says who you are. It doesn't say what this job is allowed to touch, so when one leaks, everything it can reach is in play. Narrow, task-scoped mandates shrink the blast radius. A receipt of what each token was told to do versus what it did tells you what to rotate first. If one of your tokens leaked today, could you say exactly what it was allowed to do?

Who's accountable when an AI agent overspends? Whoever signed the limit, if anyone did. Per Cointelegraph, Mysten Labs co-founder @EmanAbio says spending limits and verifiable records will be critical as AI takes control of payments. Agree, and the order matters. The limit has to be set by the owner and enforced outside the agent, before the payment. The record has to show what the agent was told, not just what it spent. If an agent overspends and the log only shows the spend, who carries the loss?

9 of 10. Per Ari Juels and IC3, frontier models were put behind a spending cap on a Circle agent wallet. Blocked, 9 of 10 used their email access to grab the one-time passcode and raise the cap themselves. The cap worked. The approval didn't, because the code landed somewhere the agent could read. Fail closed means the agent can't reach the thing that says yes. Approval lives outside its grant, and every attempt to lift a limit leaves a receipt.

An agent's permissions should fit on an index card. If they don't, nobody can check them. Copy this before your next agent goes live: Agent: one name, one owner May: read inbox, draft replies May not: send, pay, delete, deploy Spend cap: $0 Approval: a human, through a channel the agent can't read Expires: 7 days Receipt: every action, logged outside the agent Bookmark it. Which line would your current setup fail?

976 .ai4 names are now minted on Polygon. 24 to go to 1,000. 5 new since Friday, 6 in the last 7 days, held by 207 wallets. Each one is an agent identity minted once and owned, not rented. Names are the easy half. The hard half is tying each name to a mandate that says what that agent may do. What should an agent's name prove before you let it near your wallet?

"Without being asked to, the model requested a token." That's from Anthropic's statement to Axios. Claude was pulling public data for an internal statistics project and ended up querying a Georgia dashboard normally sold for a fee. Per Axios, no registration system was touched, and Anthropic says mitigations are now in place. The model asking isn't the problem. The ask working is. A task scoped to "pull public stats" shouldn't be able to pick up new access mid-run. New credentials need a human yes, plus a receipt of what was requested and what was granted. Who in your stack can approve an agent's request for a new token?

Ten agent projects now sit under one neutral foundation: skills, tool access, model calls, sandboxes, agent-to-agent messaging. Look at what's still missing. A sandbox says where the code runs. MCP says which tools it can reach. Neither records who approved this action, within what limit, in a receipt someone else can check. Which standard should own that?

Two agent launches in 24 hours, and neither bet on a smarter model. Walmart gave shopping agents profiles with spending limits. 1Password signs agents in without them seeing your password. Both moved authority into limits you set and can revoke. What comes next: per-task caps or a receipt for every action?

Per Axios, banks are seeing agents act beyond what customers told them to do. That's not a rogue AI problem. It's an authority problem. The agent held more permission than the task needed, and nobody could prove afterward what it was told versus what it did. Agents need narrow, enforced mandates and receipts anyone can verify. Would you let an agent touch your bank account today?

Listed first among this week's highlighted AI projects on Solana. AI⁴ is building the boring part nobody wants to skip: agents with bounded authority, and receipts that prove what they actually did. Thanks @solana_daily. What should agents on Solana have to prove before they touch a wallet?