Home / Articles / Who else controls AI agents at Web Summit? An outreach agent mapped eleven companies in an afternoon, in their own words, and found that most of them are partners for RiskMandate as much as competitors

Who else controls AI agents at Web Summit? An outreach agent mapped eleven companies in an afternoon, in their own words, and found that most of them are partners for RiskMandate as much as competitors

By Dinis Cruz · 2026-11-05 · article v1.0.0 · newsroom v0.14.66 · riskmandateweb-summitcompetitive-analysisagentsabpoutreachresearcharticle

Abstract: On 5 November, at a team member's ask, the Web Summit team's outreach agent and two research agents read the websites of eleven companies that work on controlling AI agents and are at, or listed around, Web Summit 2026. The result is a public page with a map, a before-during-after chain, each company's headline and offer in its own words with the source, how we know it attends, where RiskMandate differs, what we can learn from each, an honest list of where we are behind, and seven things to do before our stand on 12 November. The idea I take from it: almost everyone else works while the agent acts or after it has acted. RiskMandate works before, on the gap between what an agent can reach and what it was asked to do, so most of these products are barriers an Agent Behaviour Policy can name, and so partners as much as competitors. The newsroom checked the page against the companies' own sites: nine of the eleven headlines match where the page says, and a few details need small corrections, listed here.

In short

• Eleven companies that control AI agents, read from their own websites on 5 November by the Web Summit team's outreach agent and two research agents.
• One public page: a map, a before-during-after chain, each company in its own words with its source, and where RiskMandate differs.
• Most of them work while the agent acts or after it has acted. RiskMandate works before: what the agent can reach, what it was asked to do, the gap, and who signs it off.
• So most of them are partners as much as competitors: each can be a barrier, or the evidence, that an Agent Behaviour Policy names.
• The page is honest about where we are behind, and turns that into seven things to do before our stand on 12 November.
• The newsroom checked it against the companies' sites; a few small corrections are below.

What the page is

On 5 November, one of the people on the Web Summit team asked their outreach agent for a competitive analysis: who else at Web Summit 2026 works on controlling AI agents, how each presents itself, and where RiskMandate is different. By 23:10 UTC it was a page on the team's site: competitive analysis: who else controls AI agents at Web Summit.

What I like about how it was built:

• Each company in its own words. The headline, the offer and the call to action come from the company's own website, with the link, read on the day.
• How we know it attends, labelled. Each card says whether the company is on Web Summit's own pages (the team's company map), says so on its own site, or comes from a third-party list not confirmed by Web Summit. Six of the eleven are on the team's map from Web Summit's pages.
• Our view, marked as ours. "Where we differ" and "what we learn" are RiskMandate's view, not statements by the companies.
• A public and a private version. The public page has no threat labels and no judgements of the companies; a fuller internal version is shared only with the RiskMandate team.

Before, during, after

The page sorts everyone by the moment they work on an agent:

• Before the agent acts: RiskMandate. What the agent can reach, what it was asked to do, the gap between the two, and who signs it off.
• While it acts: Airia, Hexgate, Polygraf AI, Delinea, Arsia Labs, Descope and Keyban. Tokens, access, checks on each tool call, spending caps and filters that enforce rules.
• After it acts: Summit Cognitive, Forg3t and TrustScale. Signed decision receipts, proof that a model forgot something, and quality scores.

The sentence on the page that I would put on the stand is this one: "Most of these products can be a barrier in an Agent Behaviour Policy". An Agent Behaviour Policy lists what an agent can reach and, for each thing, the barrier that holds it to its mandate. A tool-call gate is a barrier. A payment account with caps is a barrier. A signed receipt is evidence that the mandate held. These products are the controls; RiskMandate is the record that says which controls an agent needs, and whether they are there.

The eleven, in one line each

From the team's page, with the newsroom's checks folded in:

• Summit Cognitive (closest to us): Decision Receipts, a signed record of each AI-assisted decision that can be checked offline; a design-partner pilot at a published price. It records a decision after it is made; we say what the agent may do before it acts.
• Airia (closest to us): an enterprise platform to discover, secure, govern and optimise every AI agent. A platform you buy into; we are one record per agent that starts in five minutes.
• Hexgate, by Hexamind: an open-source (MIT) layer that checks every tool call an agent makes against its rules, with an audit log; pip install hexgate. It enforces a policy; we say what the policy should be. A natural joint demo.
• Polygraf AI: a behavioural control plane between people, agents and AI systems that inspects content and enforces policy, on premises or air-gapped. It controls what goes in and out; we record what each agent is authorised to do.
• Conscium, Verifyax: tests agents in simulation for accuracy, safety, security and compliance; a free trial with no card. They test what an agent does; we write the mandate the test checks against.
• Delinea: identity security for people, machines and AI agents, with each action in a session authorised before it runs. Their record is who accessed what; ours says what the access is for.
• Arsia Labs: an open protocol for signed messages between agents, with EU rules such as GDPR, the AI Act and DORA written into it; based in Portugal. A protocol for engineers; ours is a record a business signs.
• Descope: login and authorisation for apps, AI agents and MCP servers, with scope-based access; a free tier and published prices. They issue the scopes; we decide which scopes each agent should have.
• Keyban, Agent Wallet: a payment account per agent with caps, allowed payees and human approval. Money is one line of an agent's reach; we cover all of it.
• Forg3t Protocol: makes models forget selected data on purpose, with signed before-and-after evidence. They cover the data a model holds; we cover what an agent may do.
• TrustScale: configures and measures AI output against your own criteria, with a TrustScore and expert human review. They score what an AI says; we govern what an agent can do.

Where we are behind, and what to do about it

The part of the page I value most is the column most competitive analyses leave out. Where RiskMandate is behind, in the page's words:

• No customer names on riskmandate.ai yet; several of these companies show logos or analyst mentions.
• No single headline number; several lead with usage or results figures.
• Our sample policy is not the first thing a visitor can open; Summit Cognitive's sample receipt is.
• No booking link in our outreach emails yet.
• Our hook is longer than the sharpest ones in the field.
• We don't say "free, no card" for the five-minute try.

And what the page proposes before 12 November, when our stand is at Web Summit: one number on the stand from our own sample policy (15 things it can reach, 6 it was asked to do, 7 with nothing in the way); a sample policy that opens in one tap from the stand's QR code; a booking link everywhere; one named design partner, if I agree; one sentence on each company for everyone on the stand; a joint "before, during, after" demo with Hexgate, Summit Cognitive and Keyban; and "free, no card" said plainly.

Each "what we learn" line is small and concrete: a sample anyone can open in one click, a one-line start for developers, the EU regulations named in the product itself, labels that say how each fact is known. That last one is a lesson the page already applies to itself.

What the newsroom checked

I asked for the page to be checked against the companies' own sites the same evening. Nine of the eleven headlines match on the page cited. The corrections, all small:

• Delinea: the headline quoted is on Delinea's home page, not its AI solutions page, whose heading is "Secure AI identities with runtime authorization". The 30-day trial is offered on the home page.
• Keyban: the address cited serves the French page ("Vos agents IA paient. Vous gardez le contrôle."); the English page has the quoted headline. The free offer is two wallets for three months, no card.
• Polygraf AI: its own site refused our reader, so its card rests on a third-party awards page, as the page itself says.
• Summit Cognitive: the published pilot price is $25,000 for one live workflow over eight weeks. Its site carries a banner about meeting at Web Summit Lisbon, 9 to 12 November.
• Descope: the site says scope-based access control; it does not say scopes per agent and per tool, so I have written it here without that.
• RiskMandate: the site lists one-off prices of £10, £50, £500 and £1,500 per agent; the range on the page is right, though not in those words.

Attendance is the softest part. Six companies are on Web Summit's pages as the team's company map recorded them on 1 November, and I found them in that file. Our own search did not turn up listings for any of the eleven on Web Summit's site, which says more about search than about the companies; the "third-party list" labels are the page being careful, and I would keep them.

What this does not claim

• The placements and the "where we differ" lines are RiskMandate's view, from the companies' websites, not their statements.
• Nothing here judges the companies. Several are likely partners, and the page says so.
• Attendance is as labelled, and not confirmed by Web Summit for the companies marked as a third-party list.
• The people on the Web Summit team are not named here.

Where this comes from

The Web Summit team's competitive analysis page and its data file, people/silvio/competitive/companies.json, from the commit at 23:10 UTC on 5 November 2026; the team's company map, data/companies.json. The companies' own sites, read the same evening: Summit Cognitive, Airia, Hexgate, Conscium Verifyax, Delinea, Arsia Labs, Descope, Keyban Agent Wallet, Forg3t, TrustScale, and for Polygraf AI The Hacker News awards page. RiskMandate. Earlier on this site: an analyst's blueprint for RiskMandate, selling at the right altitude, the Web Summit team, part five.

Peer review Not reviewed yet. Review it

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles