AI Governance Insights

New Jersey's A5495 is still in committee, but state disparate impact rules already expose tenant screening algorithms. What is confirmed, and what to do now.

VerifyWise 2.6 links risks across projects, vendors and models, adds retention policies to the Evidence hub, and tightens who can hold LLM keys.

VerifyWise is named in the Securing AI category of Wavestone's UK Cybersecurity Startup Radar 2026, a map of 225 UK startups across 21 domains.

VerifyWise 2.5 adds template-first reporting with schedules and run history, OpenTelemetry across the services, and frameworks moved to an extension model.

DORA did not create a new category of software to buy. It made operational resilience and ICT third-party risk legally enforceable for finance. Firms that already govern those well have most of DORA already.

The three big model risk management regimes (US SR 11-7, UK SS1/23, and Canada's OSFI E-23) were written before mainstream AI, but their technology-neutral definition of a model already covers your machine learning and LLM systems. Here is what each expects and one operating model that satisfies all three.

VerifyWise 2.4 adds three new modules: the AI Trust Index grades third-party AI apps, the AI Apps inventory governs the tools your teams use, and Agent Control governs what AI agents do.

We scored the privacy policies and terms of 205 widely used AI apps on what they disclose about training, retention, deletion and sharing, using one consistent rubric for every app. Only 23% earned an A or B, half are silent on whether your data trains their models, and one in three reserve a clause we treat as a dealbreaker. Here is what the data shows, and how the index works.

OSFI Guideline E-23 takes effect 1 May 2027 and applies to every model with non-negligible risk at every federally regulated financial institution in Canada, including AI and ML. Here is what changed, what counts as a model and how to ready your AI inventory in 18 months.

If your company puts an AI model into a regulated workflow, the question isn't only whether the model is smart. It's whether the model will refuse the request it should refuse, escalate when it should escalate and stop confidently filling gaps it shouldn't. We built a test for that and ran it across 15 models. Most of them failed.

Salary data for AI governance roles across the US, UK, Germany, France, Netherlands and Ireland, pulled from 30+ public sources. A US Chief AI Officer earns somewhere between $250K and $540K. The AIGP certification adds 13 to 27 percent, depending on what you pair it with. EU markets still trail the US by roughly a third.

The US has no federal AI law, but that doesn't mean AI is unregulated. Between executive orders, FTC enforcement, and state laws taking effect across Colorado, California, Texas, and Illinois, the compliance picture is getting complicated fast. Here's what's happening on the ground and what your organization needs to do about it.