SIRV

12/08/2026

Generative AI has an unusual failure mode: it can fail while appearing to succeed.

Traditional software usually has clearer limits. Ask it to do something it cannot do and, more often than not, the failure is obvious.

Generative AI is different.

It can produce a polished, plausible response even when the evidence is incomplete, an assumption has been presented as fact, or the underlying task has not really been completed. The interaction continues and nothing visibly breaks.

We think of this as **soft failure**.

That matters particularly in safety, security, risk and compliance, where a convincing answer is not necessarily a reliable or defensible one.

The practical question is not simply: *Did the AI produce a useful-looking answer?*

It is:

* What has actually changed?
* What evidence supports the output?
* Could the resulting decision be explained and defended later?

Our latest article looks at why fluent AI output can make failure harder to spot, and why operational AI needs controls that make missing evidence, uncertainty and weak answers more visible.

Read the full article:

Meta description:Generative AI can fail while appearing to succeed. Learn why soft failure matters in safety, security, risk and compliance, and how governance can make AI decisions more defensible.

31/07/2026

Does your contractor RAMS review end with a simple “approved”?

A document review can find missing information, conflicting details and unresolved actions. It cannot confirm that conditions at the work location match the document or authorise the work to begin.

Our new guide provides:

• a practical RAMS document-review checklist;
• three clear review outcomes;
• separate pre-start verification questions;
• common warning signs; and
• a fictional worked example.

Read the full checklist:
https://getsirv.com/2026/07/contractor-rams-review-checklist/

Use this contractor RAMS review checklist to assess documents, record actions and complete separate pre-start checks before work starts.

26/05/2026

AI can now help draft reports, summarise information, prepare briefings and support decisions.

But AI cannot carry responsibility.

After Andrew Tollinton’s guest talk to MBA candidates at the University of Sussex, one question stood out:

“My leadership team wants to ignore AI. How do I get them involved?”

Our latest article looks at why ignoring AI is not always the safe option, and how leaders can start a practical conversation about where AI is already being used, what needs human review and what should be tested carefully.

Read more: https://getsirv.com/2026/05/my-leadership-team-wants-to-ignore-ai/

15/04/2026

Martyn’s Law is not only a policy issue. It is also an operational readiness issue.

The real challenge is not simply whether a document exists. It is whether procedures are workable, site-specific and ready to use under pressure, whether plans and submissions have been reviewed properly, and whether teams can keep a clear record of what was reviewed, changed and carried forward.

That is where SIRV AI can help.

SIRV AI can support Martyn’s Law readiness by helping teams:
• retrieve the right approved procedure quickly
• review plans and submissions against expected standards
• identify gaps and weak assumptions
• support clearer compliance records
• retain lessons learned over time

In our latest article, we explain how this kind of support can help make readiness work more controlled and more usable in practice.

Read: How SIRV AI can support Martyn’s Law readiness

https://getsirv.com/2026/04/how-sirv-ai-can-support-martyns-law-readiness/

How SIRV AI can help teams retrieve procedures, review documents, identify gaps and maintain clearer readiness records.

11/04/2026

We learned from behavioural psychology how to capture attention. Now AI may be driving a bigger shift: the outsourcing of parts of reasoning itself.

Steven Shaw and Gideon Nave argue that AI is starting to act as a third way of making decisions by helping retrieve, structure and present information for us. Their warning is about cognitive surrender: handing over more judgement to AI than we realise.

The answer is not to reject AI support. It is to build the right counter around it. That is where an operational layer matters. It helps keep AI useful without letting it quietly become an unchecked substitute for judgement.

Our latest piece explores that shift, from the nudge era to the age of outsourced reasoning:https: //www.linkedin.com/pulse/we-learned-hack-attention-now-outsourcing-thought-andrew-tollinton-xm16e

31/03/2026

A lot of organisations rely on a few experienced people to spot the things others miss.

That might be a head of safety reviewing risk assessments, a security lead checking a contractor plan, or an operations manager spotting gaps in a process.

The challenge is that this know-how often stays with one person, one team or one site.

Operational memory helps organisations keep and reuse that judgement more widely.

For example, when reviewing risk assessments, experienced people often know the weak points that come up again and again:

> Vague responsibilities
> Generic risks
> Unclear emergency arrangements
> Plans that look good on paper but are hard to use in practice

If those lessons can be captured and reused, the wider team can review more consistently, newer staff can do a better first pass, and the organisation becomes less dependent on a few key people.

That is a big part of the value of operational memory.

It is not just about storing information. It is about helping teams reuse practical judgement across sites and workflows.

Read more on the SIRV website.

25/03/2026

After Andrew Tollinton, CEO of SIRV, spoke and moderated an AI panel at the Counter Terrorism and Risk Management conference at AO Arena, one thing was very clear:

Many people in security can see that AI is important, but are still unsure how to use it safely in practice.

That is a useful signal.

The gap is not really about belief in AI. It is about knowing where it fits into real security work, where it is safe to start, and how to use it without creating new problems.

In our view, security teams should not start by asking AI to make major decisions. They should start with practical support, such as:
• Finding the right procedure quickly
• Triaging incoming reports
• Drafting summaries for review
• Checking plans for gaps
• Carrying lessons learned forward

AI in security should begin with support, not blind trust.

That is exactly why controlled use, approved sources, human review and clear audit trails matter so much in high-consequence environments.

23/03/2026

Our CEO, Andrew Tollinton, will be speaking at the Counter Terrorism Risk Management Conference 2026 at AO Arena, Manchester, on 24–25 March 2026.

He’ll be speaking about AI applied to counter-terrorism risk management, and specifically how SIRV AI supports protective security teams through an operational layer. That means controlled workflows, assurance checks, evidence trails and human approval points, so outputs remain traceable, audit-ready and defensible when decisions matter.

Event details and tickets: https://www.eventbrite.co.uk/e/counter-terrorism-risk-management-conference-2026-tickets-1981567842760

If you’re attending, please say hello.

A practitioner-led Counter Terrorism Risk Management conference for those responsible for security in venues & publicly accessible places.