“Have you seen these photos of [a politician] in hospital?” a friend asked me recently, before turning his phone to me. It was a set of deepfakes from a dubious page. Not convincing if you looked closely enough but shared so many times that most weren’t looking closely.
This is no longer a political problem; disinformation costs the global economy $78 billion every year. False stories travel six times faster than the truth and reach up to 100,000 people, while the truth rarely spreads past 1,000. For the most popular brands, a single viral hoax can cause a 16% drop in reputation that fact-checking alone can’t fix.
The reputational damage is harder to quantify than the financial loss, and in many ways harder to recover from. What makes it worse is that launching a campaign has never been cheaper or easier. Bot farms are becoming big business and are used for generating revenue as much as they are for manipulation of elections.
$10 is enough to buy 50,000 fake views, 20,000 fake likes, and 500 fake comments. That’s enough to manufacture the appearance of a crisis, push a false narrative into the mainstream, and trigger a response from a communications team chasing a problem that isn’t real.
These attacks take many forms, and the most damaging are often the most subtle. A forged Department of Defense memo requesting a review of Broadcom’s acquisition of CA Technologies caused the company’s stock to drop. A fabricated tweet impersonating an aviation news account claimed Lufthansa had failed mandatory safety inspections, triggering investor panic and forcing the airline to issue corrective statements in five languages.
Disinformation campaigns have become more sophisticated and easier to deploy at a scale unseen before thanks to generative AI. These campaigns often sync with earnings reports, amplify false narratives, and create damage that spreads faster than the truth can.
This is the problem that Tom Garnett and Vlad Galu set out to solve when founding Refute. The platform they’ve built tracks narratives across social media in real time and identifies whether accounts are authentic and if the activity looks coordinated. They catch campaigns early enough to be proactive rather than reactive.
During Romania’s recent elections, Refute identified over 32,500 inauthentic TikTok videos that pushed a populist candidate. In the mining sector, it found that around 40% of sites were subject to coordinated attacks. We led Refute’s £5m seed round earlier this year, having backed them from the start, because this space is starting to look a lot like cybersecurity did a decade ago.
My friend who showed me those photos of a politician didn’t share them. He was sceptical enough to look twice. Disinformation is becoming so sophisticated and repeatable that looking twice soon might not be enough. Those photos were clumsy, but the campaigns being run against businesses right now, certainly are not.
Read on for more news from our portfolio and partners across Intelligence, Human, and Planet this month.
Intelligence
Europe’s AI edge takes a different path. In Sifted, John Thornhill explores how Europe’s AI advantage may come less from matching Silicon Valley on scale and compute, and more from building systems rooted in trust, reliability, and real-world use. Unlikely AI features as part of that shift, with its neurosymbolic approach tackling the challenge of making AI more grounded and dependable. Read the feature.
Two portfolio companies make Sifted’s compute watchlist. Sifted’s latest list of future of compute startups to watch includes both iPronics and Nu Quantum, highlighting two critical parts of next-generation infrastructure. Read the feature.
Photonic closes a major round. Photonic closes more than $200m USD in fresh funding, giving the company a $2bn post-money valuation and bringing total capital raised to over $350m USD. The milestone marks strong momentum for its distributed quantum computing architecture, built to help quantum systems connect and scale beyond a single machine. Read the press release.
Secure Agentics launches Adrian. Secure Agentics launches Adrian, an open-source runtime security toolkit designed to monitor and control AI agents before they act. The launch addresses a growing gap in agent security, where systems need oversight not just for outputs, but for reasoning, planning, and tool use. Try it for free.
Five portfolio companies feature in Barclays Eagle Labs’ AI report. PIMLOC, PolyAI, Riverlane, Secondmind, and UnlikelyAI all feature in the Barclays Eagle Labs Ones to Watch AI: 100 report. The recognition reflects the breadth of the UK AI ecosystem and the strength of companies building across privacy, conversational AI, quantum error correction, decision-making, and next-generation AI software. Read the report.
Refute features in The Times on corporate disinformation. In The Times, Katie Prescott examines how corporate disinformation is becoming more sophisticated and more damaging, driven by generative AI and fast-moving platforms. Refute features in the piece for its work helping organisations detect and respond to manipulation before it causes lasting harm. Read the feature.
PolyAI opens its platform to more builders. The launch gives enterprises faster ways to build, test, and deploy dialog agents across teams. External coverage said the release made enterprise-grade conversational AI self-serve for any team with an email address. Read the announcement.
Human
Nuclera launches an antibody triage service. Nuclera launched an antibody triage service aimed at helping teams move faster from AI-generated candidates to experimental binding data. The company said the service was designed to eliminate non-binders earlier, before more costly downstream workflows. Read the announcement.
Quibim joined New York’s strategic AI programme. Quibim was selected as the only healthtech company in New York’s strategic AI programme, marking a notable step as medical imaging AI moved further into mainstream healthcare infrastructure. The selection reinforced Quibim’s growing profile in AI-powered diagnostics. Read the announcement.
Constructive Bio’s CEO discusses commercialising the expanded genetic code. In SynBioBeta, CEO Ola Wlodek argued that Constructive Bio had moved beyond proof of concept and into the harder work of turning recoded cells into a commercial pharmaceutical platform. Read the interview.
Planet
James Baker looks at who captures value in space tech. In our latest blog, Principal James Baker examines why the winners in the space economy are not always the companies with the most elegant engineering, but those that own the system, customer relationship, and outcome. The piece offers a sharp view on what matters most for early-stage investors in space. Read the blog.
News from the team
Trailblazers and unicorns. At our London office, Stan Boland and Craig Marshall join Dr Manjari Chandran-Ramesh for the first Trailblazers Community event of 2026 to discuss what it takes to build towards breakout scale from day one. Drawing on decades of company building, leadership, and exits, the conversation centred on innovative technology, conviction, and strong leadership teams. Read the post.
If you enjoyed reading this issue of Capitalise, why not subscribe below to receive it directly to your inbox.