The same AI security defense is winning and losing — depending on which regulator hears it
From Rome to Washington, the claim that systems weren't built for AI agents was rejected when platforms raised it and accepted when banks did — decided by regulatory identity, not the merits of the argument.
Meta's defense in Rome was never that rival chatbots were dangerous — it was that they were too much.
The emergence of AI chatbots on our Business API put a strain on our systems that they were not designed to support. — Meta
Italy's competition authority disagreed, and in December 2025 ordered Meta to suspend the terms that kept competing AI chatbots off the WhatsApp Business API [1]. The argument lost again, a regulator over. Google's defense of Android's walls to the European Commission was the same shape: mandatory access, it argued, was a security risk.
This unwarranted intervention would strip away that autonomy, mandate access to sensitive hardware and device permissions; unnecessarily driving up costs while undermining critical privacy and security protections for European users. — Clare Kelly
The Commission drafted the measures anyway, demanding equal access for rival agents to voice activation, gestures, and deep tasks [2]. The same argument traveled through the rest of 2026, and everywhere it was raised by someone other than a platform in a competition docket, the wall held. In November, Amazon blocked Meta's Muse agent after it generated 29.1 million web requests in 12 days without ever identifying itself as a bot, nine percent of them touching login forms and payment pages [3]. The company Rome ordered to open was, nearly a year later, itself turned away at Amazon's door — kept out by the same claim of strain it had lost with in Europe. No competition authority has come for that wall. It held. The banks built theirs in a single late-October wave, and none came down. JPMorgan Chase, Bank of America, Morgan Stanley and Citi restricted third-party agents operating inside customer accounts [4]; RBC, TD and BMO banned them from authenticated properties outright [5]. The banks named the mechanism in physical terms.
To protect our clients, BMO does not permit AI agents to access our digital properties or execute transactions. — Bank of Montreal
It is a distinction between what an agent is allowed to see and what it is allowed to do. Meta's former security chief, Alex Stamos, put the dilemma in November: the incoming agents pass every identity check — they are not stolen credentials — yet are plainly bots [6]. Signal blocked screenshots by default because agents read the screen directly rather than the app's protected data [7]. Maybank is keeping agents off customer-facing channels until their actions can be audited end to end [8]. Every one of those walls stood. Competition authorities and prudential regulators are built to hear the same sentence differently. A competition authority exists to decide whether a refusal to deal is exclusion, and a capacity defense is, in that frame, exactly what it was built to distrust. A prudential regulator exists to decide whether a practice threatens safety and soundness, and in that frame the same words are prudence. The OCC never had a wall to test: it answered the same fact pattern by placing autonomous payment agents in its highest risk tier for bank AI evaluations [9]. The wall was neither torn down nor upheld — it was ratified, a tier higher. The vocabulary traveled further than the walls. Anthropic blocked third-party harnesses like OpenClaw from Claude subscriptions on capacity grounds [10], and analysts framed the platform blocks as moats protecting customer relationships rather than pure containment [11]. The pressure runs the other way too: 54 percent of surveyed U.S. finance leaders said they would seek workarounds if their banks refused to support agents [9]. That is a live strain on the bank side, not a forecast. The OCC is the one regulator that took the security argument at full strength and answered it on substance — and its answer was to raise the wall a tier. In Rome and Brussels the same words read as pretext; in Washington they read as findings. Meta could not keep rival agents out of WhatsApp, and Amazon could keep Meta out of its store, and the banks could keep agents out of their customers' money. The deciding variable was never the argument. It was the room the argument was made in.
- 1. Italy Orders Meta to Restore Rival AI Chatbot Access to WhatsApp
- 2. EU Orders Google to Open Android to Rival AI
- 3. Amazon Blocks Meta Muse AI Over Bot Disclosure Dispute
- 4. Major US Banks Block Autonomous AI Agents in Accounts
- 5. Major Canadian Banks Block Third-Party AI Agents
- 6. Security Experts Warn Against AI Agent Financial Access
- 7. Signal Blocks Screenshots to Stop AI Agents Reading Chats
- 8. Maybank Invests $2.5 Billion in Cautious AI Rollout
- 9. Global Banks Clash Over Security of AI Agentic Commerce
- 10. Anthropic Blocks Claude Subscription Access for OpenClaw and Third-Party Tools
- 11. Major Platforms Block External AI Agents to Protect Business