Horizon3 CTO: Attackers Are Winning With Old Bugs, Not New AI Exploits
The most alarming thing the current wave of AI-enabled cyberattacks is how boring the entry points are. Anthony Pillitiere, co-founder and CTO of security firm Horizon3.ai, says the first reported agentic threat actor — one he refers to as "Jade Puffer," which he says he learned of in July — got its initial foothold through an externally exposed Jenkins CI/CD server, a system that should never have faced the public internet, by exploiting a vulnerability from 2025. No exotic new technique. No zero-day. Just a known weakness nobody had closed, according to remarks he made on the MTS podcast.
That framing runs against the way most people imagine an AI-powered attacker. The novelty, Pillitiere argues, is volume and cost — and both are moving in the attacker's favor.
The vulnerability pipeline is outrunning the patch line
Pillitiere's central data point is supply. Vulnerabilities added to the National Vulnerability Database, the industry's running ledger of disclosed software flaws, have been arriving at almost three times the prior weekly rate. "It is significantly increasing and threat actors are able to take those very very quickly and turn those into weaponization and exploit them because organizations are not patching fast enough," he said.
The second half of that table is where Pillitiere departs from the usual security-industry script. Stopping the break-in is not the whole game, and arguably not even the main one. "The breach and the exploitation is not the main effort," he said. "You also have to understand what happens after that when there is exploitation. What does that lead to on the inside of the environment?" In other words, a company can pass a penetration test and still not know what an intruder would do once inside — which systems they'd reach, which services would go dark, and what that downtime costs.
He ties this to a familiar security framework, the CIA triad of confidentiality, integrity and availability, and to a shift in how executives are weighing it. Availability is where cybersecurity and IT operations overlap, and leadership is now treating the cost of inaction as the bigger number. "The risk of not doing something is higher than the risk of doing something," he said.
Unlimited interns, one win
The economics of an agentic attacker are what make the volume argument bite. Pillitiere's metaphor is deliberately blunt: "They have unlimited cyber interns that they get to use however they want and they can throw up a thousand of them. If 999 of them fail, but one of them passes and one of them gets ransomware, then they win."
Known, externally exposed weakness
Agentic actors probe at scale
Initial access, e.g. exposed Jenkins server
Post-exploitation impact on the business
Availability loss and disruption
Leadership reprioritizes fixes
Autonomous pen testing to find gaps first
He points to a concrete tool in that pipeline: an open-source pentesting harness called Artex, which a penetration tester published to speed up offensive workflows and which threat actors then picked up and ran at scale. Pillitiere says the FBI has issued a flash warning it. The uncomfortable implication is that defensive tooling and offensive tooling are often the same tooling — the defender's productivity gain is the attacker's too, unless the defender moves first.
Which leads to his most contrarian point, and the one most likely to irritate anyone selling an AI-security cure-all. The fundamentals have not changed. "The foundations of cyber security and the things that we've been talking for many many many years — layers of defense, asset management, understanding your multifactor authentication — all of these things are still fundamental," he said. AI does not repeal them; it raises the speed at which they must be executed. The edge goes to whoever identifies and remediates faster, not to whoever buys the newest category of product.
When the lights go out, is it an attack or an overdue repair?
Asked the recent run of high-profile outages that officials said showed no evidence of a cyberattack, Pillitiere takes the account at face value — and calls it good news. In his reading, organizations are finally addressing technical and "cyber debt," and the business impact of those fixes is now visible on the outside. He says he tried to build a theory linking the outages to attacks and couldn't, given statements from the DNI, the FBI and the affected organizations themselves.
The trade-off he describes is the same one from the CIA-triad argument, just seen from the other side: the risk of doing the fixes — planned downtime, disruption, cost — is now lower than the risk of skipping them. That is a meaningful change in corporate behavior, because technical debt in security is usually invisible until it isn't. The bill for a decade of deferred patching and unglamorous infrastructure work comes due either as a controlled maintenance window or as an uncontrolled outage. Increasingly, leadership is choosing the scheduled version.
Selling continuous offense as a defense
Horizon3's answer to all of this is a product called NodeZero, which the company markets as "autonomous offensive security" — penetration testing that runs as often, and from as many attacker perspectives, as a customer wants, rather than as a scoped, budget-limited manual engagement aimed only at the crown jewels. Pillitiere's pitch is that a single flat, outside-in view of a network is not enough; flipping the perspective to see the environment the way an intruder would, from multiple angles, produces a more granular risk picture — and one that can be tied to business impact rather than a severity score.
The market response surprised the company's own backers. Pillitiere says early investors and outside observers assumed the buyer would be enterprise red teams, the small club of large companies that already run offensive security programs. Instead, he says, demand showed up across the market — from hyperscalers with hundreds of millions of daily active users, where the value is assessing a large of the environment at scale and connecting weaknesses across it, all the way down to small businesses. The finding fits his broader thesis: if attackers need only one gap, then everyone with an internet-facing system has the same problem, just at different sizes.
Consumers are exposed, but the money is elsewhere
If the attacker economics are so favorable, why aren't individuals the primary target? Pillitiere's answer is financial: targeting a person only pays off if that person is a Jensen Huang or a Mark Zuckerberg. The approaches differ and the rewards are smaller. Most threat actors, being financially motivated, will chase the easier and larger payouts that organizations offer.
He does not minimize the exposure. He concedes real risk for older and less technical people, using his own mother as an example — he holds all her passwords and does not give them to her, because she is highly susceptible to scams. It's a small, human detail that lands harder than any statistic: the person most likely to be defrauded is often the one whose credentials someone else has to manage.
Twenty-one years in the "black world"
Pillitiere's perspective is not that of a Silicon Valley founder who discovered security as a market. He spent 21 years in the Air Force, mostly in what he calls "black world" organizations — selectively manned units with highly sensitive missions — and served as deputy chief technology officer for Joint Special Operations Command, with cybersecurity among his responsibilities. Building defensible network architectures led him to study what he was defending against. Horizon3, founded in 2019, grew out of two persistent frustrations from his time as a leader, IT operator and defender: prioritization, and translating risk into business terms.
He cites two co-founder credentials as differentiators against younger Valley founders: Snigh's prior roles — CIO at Greylock, CTO of Splunk and the first CTO of JSOC, where the two met — and the company's hiring of former US government special-operations cyber operators for exploit development and attack content, alongside architects with roughly 20 years of Silicon Valley experience. The pitch is essentially special-operations methodology applied to commercial security: assume the adversary is already thinking three moves ahead, and rehearse the attack before someone runs it for real.
The tension worth tracking
The unresolved contradiction in Pillitiere's argument is between his confidence that the recent outages are self-inflicted maintenance and his warning that agentic actors are scaling up. Both can be true at once, but only one of them is testable in the near term. The first test is whether the weekly NVD volume he cites keeps climbing, and whether the FBI's warnings harnesses like Artex translate into more post-compromise incidents at organizations that never faced a red team. The second is his economic bet that attackers will keep preferring organizations over individuals — a bet that gets tested by whether consumer-facing scam volume rises alongside enterprise breach counts, or stays flat.
The other item worth watching is Horizon3's own claim set. Hyperscaler customers, unlimited test frequency and impact-tied risk scoring are all presented entirely through its CTO, without independent verification in the episode. That does not make them wrong, but investors evaluating the security sector should treat the company's customer mix and its "everyone needs this" market thesis as management commentary rather than audited fact.
For anyone with money in software, cloud infrastructure or security, the practical takeaway is less a new product category and more a repricing of neglected basics. If Pillitiere is right that known, unpatched systems remain the practical attack surface, then the companies that will fare best are not necessarily the ones buying the most advanced AI defenses — they're the ones finally fixing the Jenkins server that should never have faced the internet. That is a boring answer, and it is exactly the kind of boring that gets expensive when ignored. The corollary for the security trade is that demand for continuous testing and remediation is likely to be driven less by fear of novel AI attacks than by the unglamorous arithmetic of accumulated debt coming due.
S&P 500 Falls for Seventh Straight Session, Enters Correction as AI Stock Rally Loses Steam
Masayoshi Son Admits Mistake: SoftBank's OpenAI Bet Cost It Anthropic
Nadella Warns AI Is Making Network Infrastructure More Fragile, With Many Systems 'Unpatchable'
Anduril's Palmer Luckey: "AI Hacking Has Outpaced Defenses... It's Time to Run"
Trust Wallet CEO Warns Banks Can't Match Machine-Speed AI Hackers
Once added, BigGo Finance appears first in Google Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
