Cantina's Hari Mulackal: AI has turned hacking into a $10 factory

Hari Mulackal wants to correct a misconception the wave of breaches that has dominated headlines this October. The crisis is not new, and it is not primarily any single hack. It is a structural change in the economics of breaking into computer systems — a change that has been building for roughly 18 months and is now spilling into public view.

"When we look at the numbers, this has been happening for the last 18 months," the Cantina Security CEO said on the Bankless podcast. "A lot of these incidents don't get reported. It's just that a couple of them, like the PHP incidents, some of the healthcare stuff, actually made the news."

The distinction matters for anyone trying to work out what happens . If October is the peak of a temporary spike, the correct response is to wait it out. If it is the moment a long-building trend became visible, the correct response is to rethink where your money and data sit.

Mulackal and Odysseas Lamtzidis, CEO of Phylax, spend most of their conversation on the second interpretation. Both run security startups with operational visibility — Cantina sells an offensive security agent called Apex and counts Coinbase, Robinhood Crypto and Mastercard among its customers; Phylax protects the Linea ecosystem and is preparing a new network launch. Their argument is that hacking has become a factory business with near-zero marginal cost, and that the defenders have not caught up.

From bespoke burglary to an assembly line

Historically, breaching a company was artisanal work. You needed talented people, weeks or months of effort, and — crucially — a profit at the end to justify the expense. That last condition was the natural brake on cybercrime. If the money wasn't there, the attack didn't happen.

Mulackal argues that brake has failed. "Hacking has now become scalable where you can build a factory where you can give it a website... and we'll go hack it," he said. "You don't even have to think costs because it will cost like $10 as opposed to like two weeks of human time."

The numbers behind that claim are striking. By Cantina's internal measure, capabilities that existed in January 2026 are now 15 times cheaper, and Mulackal expects costs to fall another 10x every six to 12 months.

Lamtzidis supplies the consequence for target selection. Crypto was once the obvious mark because the payoff at the end was clear, while Web2 monetization was murkier. At near-zero cost, that calculation disappears. "The attacker doesn't need to answer the monetization question up front anymore," he said.

Mulackal describes the resulting business model as extortion at scale: breach a database, or threaten to, and demand payment. The same actors who ran high-volume phishing campaigns are migrating to zero-day exploitation because the economics now support it.

The panic is lagging the trend

Both guests push back hard on the idea that the past week is unprecedented. Mulackal dates the capability inflection to around January 2026, when models became good enough at cyber offense and cheap enough to fire at many companies simultaneously. Cantina began running large-scale disclosure campaigns this year; the findings are only now surfacing publicly.

Lamtzidis frames the shift as the death of security through obscurity. Last year's wave was targeted social engineering and deepfakes — expensive, bespoke, aimed at specific people. The new wave is cheap, broad scanning of the entire internet.

"Suddenly crypto was always the dark forest. And we grew up in that environment," he said, referring to the industry's adversarial origins. "But the internet was never a dark forest. Security through obscurity worked."

The disclosure surge is visible in the vulnerability data. Mulackal describes a chart of cumulative published CVEs that sits mostly flat, then slopes upward, then goes vertical. He attributes the recent jump to initiatives started roughly six months earlier and warns the chart understates reality because of disclosure lag.

The mechanism behind the vertical section is what Mulackal calls the "zero-day clock" — and it is now negative. In the past, a company might have months or years to patch a vulnerability before anyone exploited it. Today the vulnerability is exploitable before the patch ships.

Lamtzidis explains why. The moment a flaw is disclosed, anyone can point an agent at the affected software and find it. There is no window for defenders to upgrade first.

Vulnerability exists in software

AI agent finds it cheaply

Exploitation at near-zero cost

Disclosure or extortion

Attacker already moved on

Defender self-attacks first

Fix findings, raise attack cost

Attacker skips target

That asymmetry is why Mulackal says companies holding customer data, reputation or money are "living on borrowed time" and needed to act six months ago. The best available moment, he argues, is today — and failing to act is a breach of duty to customers.

Offense wins for 12 to 18 months — then what?

Mulackal is blunt the near-term outlook. "Unfortunately, for the , I would say, 12 to 18 months, it's going to be largely dominated by offense," he said. "So the best way to understand what you need to do is to attack yourself."

The defensive tools exist. The bottleneck is organizational. Lamtzidis notes that lean startups like Phylax can adopt offensive-testing agents immediately, while large institutions are slowed by bureaucracy and process.

"You have a tanker, a huge ship and you have to turn it while you're being attacked by small speed boats," he said.

That image explains why adaptation speed, not tool availability, will determine who gets breached. Startups can move in weeks; banks and telcos may take years.

Lamtzidis also reframes crypto's history. The industry was mocked for weekly $100 million hacks in 2020–2021, but the whole software industry was equally insecure; crypto simply had high incentives and public visibility. Mulackal's summary is that the rest of the world is now getting the crypto industry's 2020 experience: a major incident every week is the new normal.

The silver lining both guests offer is conditional. Lamtzidis expects a "bloodbath" followed by a return to more balanced attacker-defender combat. Mulackal expects software to be materially safer in 18 months — but warns that persistent AI attackers create a world where one mistake is fatal, a world the internet was never designed for.

Crypto's adversarial upbringing is a real but limited edge

Host David Hoffman's instinct is that this environment is bullish for crypto, because blockchain systems were born under adversarial pressure. Mulackal partly agrees: firms that have operated under intense offensive pressure for years now hold an operational advantage as the rest of the internet faces the same conditions. His own company's history is securing money, which he calls the hottest target short of government-to-government cyber.

On Coinbase specifically — a Cantina customer — Mulackal says its incentive to defend has been high for far longer than a traditional bank's, so it is likely better prepared than a large financial institution. He cites North Korean actors exploiting ATMs and Target gift cards to show that legacy finance is not immune just because transactions can be clawed back. His conclusion: funds at Coinbase are not weaker today than a month ago, but the company must keep investing.

Lamtzidis offers a counterpoint that will surprise crypto natives: for the 12 months, individuals may be better off not self-custodying.

"Maybe for the foreseeable future, consider not self-custodying," he said. "Actually, maybe it's worth breaking up your assets into multiple exchanges, where not only you delegate the security to a professional team, but also they have cyber insurance."

The advice has a shelf life. It is a tactical recommendation for a specific window, not a permanent philosophy.

The thing to break may be cyber insurance

Lamtzidis flags a data point he finds genuinely alarming. In 2024, Lloyd's — which he describes as the biggest carrier in the world — warned that large language models would increase cyber risk. Yet cyber insurance premiums have fallen consistently since 2024.

That divergence between risk and price tells him carriers are mispricing exposure. He expects many to fail over the coming months.

If that happens, insurance stops working as a backstop for custodial risk just as individuals are being told to rely on it. The second-order effects are what worry him most. He points to the Ledger database leak from around 2021, which was linked to wrench attacks: attackers identified high-net-worth individuals from leaked data and went after them physically.

Mulackal extends the harm model beyond money to data and reputation, and raises the political stakes. A leak at the scale of JP Morgan, he says, would have second- and third-order effects that could reshape a country's politics.

The personal checklist

Mulackal's top recommendation is an audit. List every product that holds your data or money, ask each provider what it has changed since the crisis began, and leave if the answer is nothing. He calls it "almost irresponsible" for companies not to act now — and insists the burden belongs with builders, not users.

"If every individual is responsible for securing the software that they're using, that's not how it works," he said. "Most people are not technical. So the responsibility lies in the builders."

Lamtzidis's framework is redundancy and blast radius. "For me it's all redundancies and blast radius," he said. "Assume you are going to be exposed to an incident... have a lot of redundancies such that a lot of organizations have to fail at the same time in order for you to lose money."

In practice that means stacking self-custody defenses — a hardware wallet, a key secured by a phone's biometrics, and a separate transaction path — so a supply-chain attack on any single layer is not enough. He concedes there is no good answer for non-technical users.

Mulackal's pragmatic tip is to stop clicking sketchy links, since most escalations need a user trigger. He also urges customers to pressure companies publicly.

The cost-raising principle is the most portable idea in the conversation. "If the cost to attack you is like a dollar, like you will get attacked," Mulackal said. "Like make it like $10,000. And most people like would not attack you."

Hoffman closes by asking custodians to make proactive public statements their security posture, naming Coinbase and Infinex as examples of the transparency he wants to see.

The unresolved tension is between the guests' optimism safer software after the transition and Mulackal's warning that a world of persistent AI attackers is fundamentally different from the one the internet was built for. For anyone with money on the line, the developments worth tracking are the CVE disclosure rate over the two quarters, whether cyber insurance carriers begin to fail as Lamtzidis predicts, and whether large custodians publish concrete commitments rather than reassurance.

Nikkei Falls 407 Yen as AI-Linked Selling Leads Early Decline, Losses Trimmed to Close in 64,900 Range

US Stocks: All Three Major Indexes Rise as SpaceX and Super Micro Earnings Lift AI Shares