AI is already in use. Your policies are not yet checked on every request, call and message.
Guard. AI that designs your controls, then checks the work against your rules.
Guard uses AI agents to design the systems, controls and process around your AI, and to check them against the regulations, your own policies and outside policies such as customer and insurer requirements. Anything that does not align is blocked, challenged or sent to a named person to approve. Guard and Evidence are built as a pair: Guard acts, Evidence keeps the proof.
A rule on paper does not stop a prompt, a call or a message.
of UK employers have a formal policy on generative AI at all.
CIPD, 2026 (survey)People are told what to do, but nothing checks it when a prompt is sent, a document is generated or a customer is answered.
Contact-centre calls and customer messages carry the most regulatory risk, and are usually sampled by hand, if at all.
When something does not align, there is no clear route to a person with the authority to approve, challenge or stop it.
Rules, tools, checks and limits around AI, so it behaves as intended.
A harness is the set of rules, tools, checks and limits wrapped around an AI model or agent so that it behaves as intended. It decides what AI may see, do and say, checks each output against your rules, sends anything doubtful to a named person and keeps a record. GenWA does not build the AI harness. We are harness deployment and implementation specialists and partners: we apply and run best-practice harnesses, from design to use to deployment. The GenWA Toolkit includes harnesses that supplement your own.
A harness is not a single tool. Guard acts; Evidence keeps the proof.
We design, run and put harnesses in place, fitted to your business.
We work out where AI touches your systems, which policies and regulations apply, and where a person must approve. We start from best-practice harnesses, AI agents draft the controls and process, and your owners confirm them.
Every day the harness checks documents, prompts, calls and messages against your rules. It can block, challenge, flag for approval and report. Your people see why, not just that.
The harness runs in your AWS or Azure account, or hosted by GenWA on AWS as agreed. Controls apply to the AI you connect and route through GenWA. Personal data is redacted before approved outside model calls.
Documents are produced as AI is used, then flagged to a person to approve.
When AI helps draft a document, Guard generates the supporting record alongside it: what was asked, what rules applied and what was produced. Anything consequential is flagged to the person with the authority to sign it off.
Guard blocks, challenges and escalates what does not align with your rules.
A plain report of what aligned with your policies and the rules you selected, and what did not.
Staff and customers can question a flagged result. The challenge, the decision and the reason are kept.
Where a rule says stop, Guard can block the action before it happens and tell the person why.
Consequential items go to the person whose authority covers them. No one else can approve.
If your contact centre uses Amazon Connect, we can listen to calls, guard them and police them against your policies and regulations.
We analyse your customer messages and report on anything that does not align with your policies and the external rules and laws you select.
Estimate how much checking Guard could take off your team.
Enter how many calls, messages and documents your people check by hand, and how long a check takes. The calculator shows the hours involved today, and how they change if only the items Guard flags go to a person. It is an illustration from the figures and assumptions you enter, not a quote, a forecast or a promise of saving.
Review workload calculator
Workload estimate, from the numbers you enter
- Every output that needs review is reviewed once, by one person.
- Who must review, and what counts as consequential, is your decision, not an estimate.
Items marked assumption can be edited. No figures are stored or sent.
How this is worked out
Reviews a week = outputs a week x share needing review. Review hours = reviews x minutes per review / 60. Share of one person's week = review hours / working hours in a week.
Guard reads, reviews, flags and reports, so owners see only what needs a decision.
Guard reads the material, reviews it against the rules, flags what does not align, and reports to the person who owns it. Reports feed Evidence, so the same findings become the record you can show.
- Sees the AI you connect and route through GenWA, in one place.
- Holds your policies, the regulations you select and the outside rules that apply.
- Decides what is allowed, what is blocked and who must approve.
- Sends every exception to the person with the authority to decide.
- Your domain expertise, blended and wrapped by our AI knowledge and know-how, which keeps evolving and improving.
- Best-practice AI skills and behaviours that we load into your agentic organisation.
- Guard products: Evidence packs and the GenWA Toolkit, whose harnesses supplement your own.
- Built with your policies and your business intelligence, kept current by us.
GenWA frame is our control framework and the foundation of the GenWA Toolkit. It joins your policies and business intelligence and combines them with our technical expertise. Guard and Evidence draw on it. Parts of the frame are available today, including Evidence Packs and Compliance Mapping. The rest, including the shared library of skills and plain-English rule writing, is being built, and we will tell you plainly which is which. An agentic organisation is the set of AI agents that do work in your business, with named people accountable. Controls apply to AI used through GenWA.
Guard is shaped to the size and structure of your business.
Choose the tab that sounds most like your business.
Legal and compliance skills without a law-firm retainer
You rarely have an in-house compliance team, and a standing retainer is hard to justify. Guard brings that discipline into the business cost-effectively: your policies and the rules you select are checked as work happens, and the owner sees only what needs a decision. It supports your compliance work and does not replace legal advice where you need it.
See what suits small businessesRules that hold when the owner is not in the room
In an owner-managed business, decisions often sit with one or two people. Guard writes down who may approve what and checks AI work against it, so sign-off does not wait for the owner and nothing consequential slips through when they are away.
Controls that keep pace with growth
Customers, insurers and investors start asking how you control AI just as the team is changing fast. Guard sets the rules once and applies them as new people, tools and teams arrive, so growth does not outrun your controls.
One rulebook across many teams
Each department uses AI differently and the risk team is small. Guard gives every team a harness shaped to its work under one shared rulebook, and sends exceptions to the right department head rather than to the central risk team.
See what suits mid-market firmsNamed people stay accountable, and the checks show it
Where senior managers answer to a regulator, an AI policy is not enough. Guard checks calls, messages and documents against the regulations you select, blocks or challenges what does not align, and routes consequential items to the person with authority. People decide on credit, hiring and medical matters, not AI.
Layered rules across business units
Large groups carry company-wide rules, unit rules and local rules. Guard layers them, so a team sees the rules that apply to it and the board sees one picture of blocks, challenges and approvals across the group.
See what suits enterprisesBuilt for the people who carry the risk.
Sees where AI could breach a rule, what was blocked and what is waiting for approval, in one report.
Gets the compliance report, the challenges and the sign-offs against the regulations that apply.
Keeps calls, messages and documents moving, with fewer manual checks and clear owners for exceptions.
Mapped to the standards, rules and regulators that apply to you.
- ISO/IEC 42001 (the international standard for managing AI)
- UK GDPR (UK data protection law)
- SOC 2 (an independent security audit standard)
- Cyber Essentials (the UK government-backed security scheme)
- Financial Conduct Authority Consumer Duty (its customer outcomes rules)
- National Cyber Security Centre guidance
Start with advice, run the product, or build with the incubator.
Our advisers, human and agentic, review your AI use against your policies and the rules that apply, and design the harness on paper first.
See advisoryProductGuard's products are Evidence packs and the GenWA Toolkit, sold as, whose harnesses supplement your own. They run in your AWS or Azure account, or hosted by GenWA on AWS as agreed.
See the GenWA ToolkitIncubatorWe build a harness for a workflow you want to put AI into, with controls designed in from the start.
See the incubatorKeep reading
Guard
Leapfrog your own roadmap: what APIs teach about MCP
How open connections let large businesses grow faster than their trajectory allows, and how to do it with control.
By Steven Myers · 10 min read
Guard
Harnesses: the controls that make AI safe to put to work
What a harness is, what it does, what happens without one, and why deploying it well is a skill of its own.
By Lukasz Piotrowski · 5 min read
Straight answers on Guard.
What is GenWA Guard?
GenWA Guard is a product that designs and runs the controls around your AI, and checks work against your policies and the regulations you select. It can block, challenge, flag for approval and report, and it is built to work with Evidence, which keeps the proof.
What is a harness, and do you build it?
A harness is the set of rules, tools, checks and limits wrapped around an AI model or agent so that it behaves as intended. GenWA does not build the AI harness. We are harness deployment and implementation specialists and partners: we apply and run best-practice harnesses, covering design, use and deployment.
What is GenWA frame?
GenWA frame is our control tower and knowledge base. It joins your policies and business intelligence with our technical expertise, and loads best-practice AI skills and behaviours into your agentic organisation. Parts are available today, including Evidence Packs and Compliance Mapping, and the rest is being built.
What is the GenWA Toolkit?
The GenWA Toolkit is a paid collection of our own tools, refined, maintained and upgraded, sold as. Its harnesses supplement the ones you already have. It is being built, and the proposition is described on its own page.
Can Guard monitor calls and customer messages?
If your contact centre uses Amazon Connect, we can listen to calls and check them against your policies and the regulations you select. We can also analyse your customer messages and report on anything that does not align. Anything flagged goes to a named person for review.
Will Guard make us compliant or certified?
No. Guard and Evidence are control-mapped to the frameworks we name, and we prepare the evidence. Accredited bodies issue certification, and this is not legal advice.
Is AI always the right tool?
No. Where the answer must be identical every time, or the decision is about credit, hiring or medical matters, a rule or a person is better. Part of designing a harness is choosing the right tool, and Guard helps you decide which.
See Guard on your own policies.
Control-mapped, not certified. We prepare the evidence; accredited bodies issue certification. This supports compliance and is not legal advice.