$ today_in_10_seconds
Ant International, Mastercard and Visa are building a shared Know-Your-Agent framework so an agent verified on one network is recognised on the others. In the same week, Anthropic published evidence that multi-agent systems are already running full cyber kill chains, and California made third-party AI audits law. Identity, evidence, enforcement, all three inside 72 hours.
$ lead_story
Payment networks start issuing agents an identity
Ant International, Mastercard and Visa have begun work on a Know-Your-Agent interoperability framework to let card networks, wallets, agent platforms and marketplaces onboard and identify AI agents across networks under shared principles, while each network keeps its own verification and decisioning. Three incompatible schemes already exist: Visa's Trusted Agent Protocol, Mastercard Verifiable Intent, and Ant International's Agentic Mobile Protocol. The work runs through BuildFin.ai, the industry platform convened by the Monetary Authority of Singapore. The problem it solves is boring and enormous: a merchant seeing a purchase from software rather than a person currently has no shared way to know whether the agent is real, who built it, or whether it is allowed to spend on someone's behalf. This is a standards effort, not a shipped product, and liability, revocation, data sharing and dispute handling all remain open.
// why_it_matters: Agents that cannot prove who they are get treated as bots and declined, which makes identity the gate on every dollar of agentic commerce.
$ also_worth_attention
Accenture and Google Cloud form the Gemini Enterprise Business Group
Accenture and Google Cloud launched the Accenture Gemini Enterprise Business Group on September 8, a global group aimed at helping clients scale Gemini Enterprise outcomes, anchored by a planned 1,000-person forward deployed engineer workforce and building on Accenture's roughly 50,000 Google Cloud-skilled professionals. Strip the press release and the signal is clear: the bottleneck on enterprise agent deployment is not model quality, it is the absence of people who can wire agents into messy internal systems. When the answer to an AI platform problem is a thousand engineers on client sites, the tooling is not ready yet.
link →Anthropic: attackers are running agents, not prompts
Anthropic published its September 2026 threat intelligence report on September 10, covering activity disrupted between December 2025 and August 2026 across seven harm areas including cyber operations, influence operations, surveillance and illicit distillation. A majority of documented cyber operations involved direct AI orchestration, with multi-agent systems performing tasks across the kill chain from target discovery through access, persistence and exfiltration, narrowing the gap between nation-state operators and low-resource actors. The uncomfortable read for builders: the offensive side has productionised multi-agent workflows faster than most enterprises have productionised a single one.
link →California makes independent AI audits a legal category
Governor Newsom signed two bills establishing the first US framework requiring independent third-party audits and assessments of AI systems. SB 813 creates a framework for independent verification organizations that assess AI systems for compliance with state law, and AB 1405 creates a state registry for AI auditors with standards for independence, transparency and integrity. Scope reaches past frontier labs to any company deploying an off-the-shelf model to screen job applicants or price insurance. Registration becomes mandatory for covered audits from January 1, 2029, which sounds distant until you remember audit evidence has to be generated before the audit, not during it.
link →
$ bigger_picture
The trend behind the headlines
Three unrelated organisations converged on the same missing primitive this week: proof of who an agent is and what it was allowed to do. The card networks want it so merchants can accept agent traffic. Anthropic's threat data shows what happens without it, with autonomous pipelines wearing stolen credentials and looking indistinguishable from legitimate automation. California now wants an auditor with a registration number to attest to it. None of this is about model capability, and that is the point: the 2026 constraint on agentic AI is not reasoning, it is accountability plumbing. If you are building agents that touch money, customer data or regulated decisions, the identity and logging layer you skipped in the prototype is now the thing that determines whether it ships.
Ant International Agentic Mobile Protocol (AMP)
Now open-sourced on GitHub with source code, developer SDKs and technical documentation for AI platforms, wallets, acquirers and financial institutions. The design is worth studying even if you never touch payments: users authorise a task rather than hand over an account, permission boundaries are explicit, and a Know-Your-Agent layer certifies an agent's authorised capabilities with a trust rating that gates how much autonomy it gets. First-phase rollout covers ten Alipay+ wallets and seven acquirers including Adyen, Checkout.com, Fiserv, Global Payments, Nuvei and Worldline.
$ subscribe --free
Get the next edition in your inbox
// DAIly Agent covers agentic AI three times a week
// free, no spam, exit anytime