AI For Brokers In London: The FCA’s New Transaction Reporting Regime And Its Surveillance Warning

3 Apr 2028 - Start of the new UK transaction reporting regime (FCA CP26/34, 2 October 2026)  ·  65 to 52 - Reportable fields after the reforms (FCA PS26/15, 3 August 2026)  ·  7 - Retail brokers reviewed in Market Watch 86, all required to strengthen surveillance (FCA, 14 October 2026)  ·  6 Nov - Closing date for comments on CP26/34 (FCA)

What is changing in UK transaction reporting?

Transaction reporting is the daily submission to the FCA of the details of trades in financial instruments, which the regulator uses to detect market abuse. The UK is replacing its current regime, inherited from MiFIR, with a simpler one.

The policy was set in PS26/15 on 3 August 2026. It cuts the number of fields “from 65 to 52”, removes FX derivatives from scope “reducing costs for over 400 UK firms” and shortens the back-reporting period from five years to three. The FCA said: “We expect these changes to reduce firms’ reporting costs by more than £100m annually.”

On 2 October 2026 the FCA followed with CP26/34, which turns the policy into technical detail. “We have published our draft schema and validation rules to help firms prepare for 3 April 2028,” it said, and asked for comments by 6 November 2026. A user pack consultation follows in the first quarter of 2027. A-Team Insight reported on 6 October that legacy formats will not be accepted after the switch, and that the legs of a package transaction will need a common identifier.

What did Market Watch 86 find at retail brokers?

Market Watch 86, published on 14 October 2026, is blunter. The FCA said: “We visited 7 retail brokers to assess the effectiveness of their market abuse surveillance arrangements.” Ten years after the UK Market Abuse Regulation took effect, it said, “these arrangements should be well-established.”

The findings will be familiar to anyone who has inherited a surveillance system:

• Vendor defaults. Some staff “appeared to rely on out-of-the-box settings provided by third-party vendors”.
• Thresholds that did not fit the business. Some firms used a flat price move threshold that ignored the difference in volatility between large and small companies, and some set minimum trade values so high that much client activity was never monitored.
• Alerts mistaken for controls. “Some firms mistakenly assumed that having an alert in place was evidence of an effective control.”
• Weak testing. Some firms “relied on simple testing metrics such as comparing alert volumes to trading volumes”, which missed alerts that were not triggering at all.
• Senior managers out of the loop. “In some cases, senior managers were unaware of issues identified by surveillance teams.”

The consequences were real. “We required all firms to take action to strengthen their surveillance arrangements after our visits,” the FCA said, and in some cases it used skilled person reviews. The day before, on 13 October, the FCA and the National Crime Agency announced three arrests in an operation against a suspected insider dealing ring.

Where does AI for brokers in London actually help?

Both publications describe the same underlying problem: a firm can have a process that runs every day and still not know whether it works. That is where AI, used carefully, earns its keep. An agent does the repetitive checking a team rarely has time for, and a person makes every judgement.

• Transaction report checks before submission. An agent compares the day’s reports with trade records, flags missing trades, duplicates and fields that do not match, and a person approves every correction and the submission itself.
• Testing against the draft schema. With the FCA’s validation rules published, a firm can run its current reports through them now and see which fields will fail in 2028, long before the deadline.
• Calibration evidence. An agent can prepare above and below the line samples and synthetic test trades, so analysts can show that each surveillance scenario fires when it should. The FCA named both as good practice.
• QA and MI. An agent can sample closed alerts for quality assurance and draft the monthly surveillance MI, so senior managers see what the surveillance team sees.

How should a broker build agentic AI into surveillance and reporting?

Agentic AI is useful here because the checking involves several steps across several systems: pull the trades, pull the reports, compare, explain the difference, draft the fix. The design matters more than the model. Four rules keep it safe:

1. Read-only access to trading and reporting data, with no ability to submit to the FCA or change the trade record.
2. A cited source for every flagged difference, down to the trade reference and field, so a reviewer can confirm it quickly.
3. A named approver for every correction and every submission, recorded in the log.
4. A written description of what the agent checks and what it does not, kept with the firm’s market abuse risk assessment.

This is ordinary workflow automation with a hard stop before anything leaves the firm. Our engineering guide to AI trade surveillance alert triage shows how the triage side can be built, and our article on approval gates in agent workflows covers the controls.

Will AI replace brokers or surveillance analysts?

Nothing in either FCA publication points that way. Market Watch 86 asks for better-calibrated alerts, more testing and better-informed senior managers, which means more evidence for people to judge, not fewer people judging it. The FCA’s own approach is similar: its chief executive said on 22 September 2026 that it is exploring agentic AI as a “first responder” for monitoring wholesale markets, alongside supervisory judgement.

What changes is where analysts spend their time. Less of it goes on assembling data and more on deciding what the data means.

What should a broker or trading firm do this quarter?

1. Read Market Watch 86 against your own surveillance set-up, starting with any scenario still running on vendor default settings.
2. Run last month’s transaction reports through the draft schema and validation rules published with CP26/34.
3. Decide whether to respond to CP26/34 before 6 November 2026, particularly on package transactions and data your firm finds hard to source.
4. Add surveillance testing and reporting accuracy to the monthly MI pack. Our agents can prepare those figures for MI and board packs.

If you are choosing a workflow automation agency to help, ask to see how it handles approvals and logs before you look at anything else. We set out what to ask in our guide to choosing an AI agency in London.

How can a broker test report checking on its own data?

BraivIQ’s 14-day Proof Run works read-only on a firm’s own exports, such as a day of trades and the matching transaction reports. It finds every mismatch and times the checking on real cases, so the firm has a signed baseline before anything is built.

The senior team behind BraivIQ has built real-time trade surveillance for one of the world’s largest trading floors, where routine monitoring workload fell by more than 70% and people made the call on what each alert meant.