In late July 2026, employees and leaders from OpenAI and Anthropic began petitioning the US government to slow down the development of frontier AI models[1]. When the people building a technology formally request intervention because of internal safety concerns, it signals a shift. For an advice firm owner operating under FCA rules, this is not just an industry debate. It is a material change in how you must assess the risk of the tools you use.

Why are AI developers asking for a slowdown?

The gap between how fast AI systems are being deployed and the maturity of the safety infrastructure governing them has become too wide to ignore. Leading frontier AI labs currently lack documented containment plans for rogue models, and they are openly acknowledging the risk[2].

For a regulated advice firm, this presents a distinct problem. I regularly see AI adoption expanding across the advice industry, with recent data showing 74 percent of firms using it in some capacity[3]. But as these systems move deeper into workflows, relying on third-party infrastructure that the creators admit they cannot fully contain introduces a severe operational risk. You cannot assume a tool is safe simply because it comes from a major provider.

What does an AI slowdown mean for your Consumer Duty evidence?

You can no longer rely on generic vendor assurances of safety to satisfy your regulatory obligations. The Consumer Duty requires your firm to avoid foreseeable harm to retail customers. If the leading AI developers are publicly stating that current deployment speeds outpace safety controls, any harm caused by an unmonitored AI system in your firm is by definition foreseeable.

Under the Consumer Duty, you cannot outsource the responsibility for avoiding foreseeable harm to a software vendor.

This means your compliance approach must shift from trusting the model to actively constraining it. If a system handles client data or drafts communications, you need documented proof of how you monitor its outputs and limit its autonomy. The system can flag an issue or draft a suitability report, but a qualified human must retain the final judgement and sign-off.

What should an advice firm do to manage AI risk today?

You need to move from trusting the model to actively constraining it. Here are three practical steps to help your firm manage these risks without pausing your automation projects.

First, pin your model versions. Do not use floating aliases like ‘gpt-4-latest’ in your production environment. Use concrete model slugs so you have a fixed baseline for regression testing[4]. If a model updates silently, its behaviour changes silently, which breaks your compliance evidence.

Second, adopt a glass box approach. Set up your workflows so that human agents and compliance teams can monitor and intervene in AI decisions. Every AI-assisted workflow needs a clear, auditable trail showing exactly where the system acted and where the human reviewed it.

Third, map your data retention. Start with the workflow, identify every place it creates a persistent state, and check each against the applicable provider arrangement[5]. You must know exactly where client data sits when it leaves your core systems.

The calls for an industry slowdown are a clear signal that the era of moving fast and hoping for the best is over. Regulators will expect your governance to reflect the known risks of these systems. If this is the situation your firm is in, a discovery call with Cordrey Consulting is a good place to start.

This article is for informational purposes only and does not constitute regulated financial advice or a compliance opinion. Consult a qualified compliance professional for advice specific to your firm.

Sources

[1] The Verge, ‘AI leaders petition US government to slow down AI development’, The Verge, 29 July 2026. Available at: https://www.theverge.com/ai-artificial-intelligence/972161/ai-leaders-us-government-openai-anthropic-google-meta [2] TechCrunch, ‘Frontier AI labs still won’t say how they’d contain a rogue model’, TechCrunch, 22 August 2026. Available at: https://techcrunch.com/2026/08/22/frontier-ai-labs-still-wont-say-how-theyd-contain-a-rogue-model/ [3] Intelliflo, ‘2026 Advice Efficiency Survey’, Intelliflo Insights, 24 July 2026. Available at: https://www.intelliflo.com/insights/thought-leadership/2026-advice-efficiency-survey-form [4] Digital Applied, ‘GLM latest floating alias model pinning risk’, Digital Applied, 23 August 2026. Available at: https://www.digitalapplied.com/blog/glm-latest-floating-alias-model-pinning-risk [5] Digital Applied, ‘AI agent platform data retention eligibility census’, Digital Applied, 10 September 2026. Available at: https://www.digitalapplied.com/blog/ai-agent-platform-data-retention-eligibility-census