A fleet of autonomous AI agents that hunt down impersonators, phishing kits, and deepfakes across the internet - and pull them offline in hours, not weeks.
Somewhere right now, a customer is typing a password into a page that looks exactly like your bank, your favorite app, your employer. The logo is right. The font is right. The little padlock is right. Everything is right except the part that matters: it isn't you. Multiply that by a few million, give the attackers generative AI, and you have the texture of the modern internet.
Outtake exists for that exact moment. The New York company runs a standing army of AI agents that crawl the open web, social platforms, app stores, ad libraries, and the dark web, looking for the fakes wearing your face. When they find one, they don't file a ticket. They trace the campaign behind it, classify it, and take it down - on a clock measured in hours.
It is, depending on your mood, either a deeply unglamorous job or the most interesting one in security. Outtake has decided it is both.
"We're taking down AI impersonation in hours."
- Outtake's own promise, printed in plain type on its homepageFor years, brand protection was a help-desk in disguise. A human analyst spotted a fake domain, screenshotted it, emailed a registrar, and waited. Days passed. The scam ran the whole time. By the time the takedown landed, the attacker had already spun up ten more pages from the same kit.
Then generative AI arrived and quietly broke the math. Cloning a brand used to take effort; now it takes a prompt. Deepfaked executives, fake support accounts, counterfeit apps, malicious ads - the cost of producing a convincing fake collapsed toward zero, while the cost of removing one stayed stubbornly human.
Outtake's founding observation was unsentimental: if attackers had moved to machine speed, defenders couldn't keep answering at office-hours speed. The only honest response to automated fraud is automated defense.
"The cost of faking you fell to zero. The cost of un-faking you didn't - until somebody automated it."
- The thesis, paraphrasedOuttake was founded in 2023 by Alex Dhillon, who spent his earlier career inside Palantir's experimental product team, reporting directly to CTO Shyam Sankar. That detail matters less for the resume line than for the worldview: Palantir's whole premise is that messy, sprawling real-world data can be made operational if you build the right agents on top of it.
Dhillon's bet was to point that instinct at the external internet - the unowned, hostile sprawl where impersonation lives - and let AI agents do the watching, the investigating, and the dismantling. Not a dashboard that tells you about threats. A system that removes them.
The bet attracted believers. When Dhillon raised, the people who had run security and software at the largest scale imaginable wrote checks personally. Shyam Sankar, his old boss, became an investor. So did the CEOs of Microsoft and Palo Alto Networks. Which is a strange thing to happen to a three-year-old company, unless the idea is right.
Outtake's platform collapses a workflow that used to span five tools and a dozen people into a single loop. Agents detect the threat, expose the infrastructure behind it, classify it, and execute the removal - then confirm the kill. It plugs into the SIEM and SOAR stacks security teams already run, so it adds capability instead of another silo. The work splits into a few sharp specializations.
Finds and removes fake domains, cloned apps, and malicious ads by exposing the campaign infrastructure underneath them.
Catches deepfakes, doxxing, and extortion aimed at leadership before any of it reaches the public.
Maps coordinated protests, venue targeting, and route exposure risks tied to physical places and events.
Shuts down unauthorized resale, trial abuse, and token fraud against SaaS platforms and digital goods.
An investigation agent that traces an attack early - often before it ever makes contact with its target.
A published framework breaking AI-driven attacks into eight stages, so defenders share one model for disrupting them.
Alex Dhillon leaves Palantir's experimental product orbit and starts a company built on autonomous agents for external threat defense.
Early institutional conviction that automated takedown is a category, not a feature.
The company ships an early-investigation agent and publishes its eight-stage framework for AI-driven attacks via Outtake Labs.
Satya Nadella, Nikesh Arora, Bill Ackman, Shyam Sankar, Trae Stephens, Bob McGrew, Guillermo Rauch and John Donovan join the round.
Revenue grows sixfold and the customer base more than tenfold year over year, with 20M potential attacks scanned.
Marketing claims are cheap; takedown clocks aren't. The case for Outtake rests on a handful of figures that are awkward to fake, because customers notice when a scam stays up. The chart below shows the gap the company is selling against - and where it claims to land.
Takedown confirmation rate - the share of removals Outtake verifies as actually gone.
Platforms, marketplaces, and dark web sources monitored continuously.
Potential cyberattacks scanned in the prior year by the agent fleet.
And then there is the customer list, which reads like a who's-who of organizations that cannot afford to be impersonated.
"Outtake protects AI labs from AI-generated fakes - OpenAI and Anthropic are both on the customer list."
- A sentence that would have read as science fiction in 2021The $40M Series B was led by Iconiq Capital. The angel list is the part that makes people look twice - the operators who built security and software at planetary scale, putting in personal money.
Strip away the funding headlines and Outtake's stated mission is almost old-fashioned: make the internet trustworthy again. The phrase the company uses - "restore digital trust" - sounds like a slogan until you remember what trust actually buys. It's the reason you click the link, install the app, answer the call. Erode it, and commerce gets slower and meaner for everyone.
The agentic internet, where software acts on our behalf, raises those stakes. If your agents and mine are transacting automatically, a convincing fake doesn't just trick a person; it tricks a machine that won't pause to feel suspicious. Outtake's argument is that trust at machine speed has to be defended at machine speed.
"Trust is what makes you click the link. Someone has to defend it at the speed it's being attacked."
- The mission, minus the sloganReturn to where we started: a customer, a password, a page that was never yours. In the old world, that page lived for days while a human worked the phones. In the world Outtake is building, an agent saw it appear, recognized the kit, mapped the campaign, and pulled it down before most of the intended victims ever loaded it.
That's the whole company in one image. Not a smarter alarm - a faster eraser. As fakes get cheaper and the internet gets more automated, the question won't be whether you'll be impersonated. You will. The question is how long the fake gets to live. Outtake is betting the answer keeps shrinking, and that 14.4 hours will one day sound slow.
Whether they're right is, for now, an open case. But the people who built the last era of security are betting they are - and they put their own names on it.