InMarketing Today

OFSI, the Treasury's sanctions enforcer, has fined Citibank's London branch £4.7 million for processing 970 payments worth £19.7 million to designated Russian entities, most of them in the months following Russia's invasion of Ukraine. The payments included transactions for sanctioned banks Alfa-Bank, Gazprombank and Credit Bank of Moscow. OFSI cut the penalty by 40% because Citi self-reported the bulk of the breaches and agreed a settlement.

The story is really about the gap between what Citi did and what it said. Voluntarily self-reporting the bulk of 970 breaches is a genuinely credible act — it earned that 40% discount. Yet Citi's public statement buried the fact behind boilerplate: “proactively engaged”, “pleased to conclude this matter”. The bank did the hard thing, then described it in the softest possible language.

Any regulated firm negotiating a self-reported settlement should treat the discount itself as the headline, not an afterthought. State plainly what you found, what you fixed, and why voluntary disclosure earned credit — regulators reward it, and so, quietly, does trust with clients and counterparties.

Binance has launched Agent OS, a developer platform giving AI tools such as ChatGPT, Claude Code and Cursor standardised, permissioned access to its trading, wallet and market-data systems. Users assign agents to dedicated subaccounts, set their own limits, and can revoke access at any time; Binance monitors the resulting trades but not the agent's reasoning. “Binance Agent OS addresses the fragmentation developers face when building agentic finance applications,” said Jeff Li, the exchange's VP of product.

The product here is not really the AI integration — every exchange is racing to bolt one on. The product is the permission model: subaccounts, revocable access, monitored trades. Binance is marketing control as the feature, a smart read of where trust actually breaks down when you hand a financial account to software that reasons for itself.

If your firm is building or buying AI agents with any access to money or client data, make the control layer — who can revoke what, and how fast — the first thing you explain to a client, not the last. That explanation is now a selling point, not small print.

LinkedIn added 1.4 million active users across the EU in the first half of 2026, taking its EU active base to 56.5 million, according to the platform's latest DSA transparency disclosure. The more striking number is enforcement: detected instances of inauthentic activity — engagement pods, automated posting, AI-generated “slop” — rose 46% over the same period. Active users remain roughly 30% of LinkedIn's stated EU membership, a ratio that has barely moved since 2024.

This is the gap between LinkedIn's headline numbers and what marketers should actually plan against. “Members” and “double-digit growth” make good slides; active users, flat at roughly three in ten, is the number that should set your content plan. A platform simultaneously reporting growth and ramping up its fight against AI-generated engagement is telling you plainly: reach is inflated, and increasingly, so is some of the content filling it.

Plan LinkedIn activity against active-user reality, not membership totals, and apply the same scepticism to any engagement stats a vendor or agency hands you. If a post's engagement looks too smooth, ask who — or what — is doing the liking.