Google’s Potential, AI Money Spigot Shutting?, PHP Bug & Cybersecurity’s “Black October”

Alex Stamos · Cognition

The PHP bug is hard to fix because patches cover only the latest PHP line, forcing risky code refactors.

Stamos explains the PHP remote code execution flaw is pre-authentication, so attackers need no login. The only patch targets the latest PHP family, and many sites run older versions. Upgrading breaks those sites, and the fix was not back-ported, so many companies must refactor code to get compatible with PHP 8. Unofficial back-ports exist but are less available, so exposure persists for many sites.

“And you can't fix the bug without patching PHP, which is under your web code. They only released a patch for the latest family of PHP, and there's a lot of PHP sites on older versions. And so upgrading that breaks your site.”at 41:47

Alex Stamos · Cognition

Secondary breaches will surface over the coming months as attackers use footholds from unpatched systems.

Many companies patched within days but did not realise attackers may have gained a foothold in the forty-eight hours before patching. Stamos expects breach notifications to arrive over a long period as companies with good security teams detect and investigate. He also expects secondary breaches over the following months as attackers move deeper using those footholds. Ransomware actors will ask for payment, while state actors stay quiet.

“And so then we'll see the secondary breaches over the next months, as attackers now use that foothold to go deeper and deeper inside these companies.”at 48:34

Alex Stamos · Cognition

Consumer fraud, spam and password-based scams will hit new highs over the next six months.

Stamos expects dark-web sales of tens of millions of stolen passwords. Many people reuse passwords, so those credentials will work on their email and banks. Email access then lets attackers reset other passwords. He expects a lot more individualized fraud and spam, plus AI-driven scam calls and texts, which he says will hit a new high over the next six months.

“we're gonna see a ton of kind of individualized fraud, spam, that kind of stuff is all gonna hit a new high over the next six months.”at 49:55

Alex Stamos · Cognition

Cyber threat tempo has shifted to a heightened level for several years, not returning to baseline.

Stamos argues the PHP bug and the wave of breaches in the past month mark a new normal. Attackers are running automated exploitation of hundreds of millions of attempts across millions of websites, and AI is speeding up their operations. He expects the pace may ease somewhat but will stay well above the old baseline for several years, so both attackers and defenders stay at a heightened operational tempo.

“this is probably the new normal for several years where the pace of crazy things that have happened in the last thirty days might be... This might be up here, but we're not going back to the baseline. It might drop a little bit, but it is possible we are now at a heightened level of kind of operational tempo for bad guys and for defenders for late... At least several years.”at 52:47

Alex Stamos · Cognition

Professional state-level hackers are targeting AI infrastructure providers like neoclouds and hosts for API keys.

Stamos points to breaches at AI ecosystem companies such as Fireworks and Railway. He says the attackers look professional, probably working for commercial vendors doing state-level work, not teenagers. Attackers are targeting important AI-ecosystem players like neoclouds and hosting providers, because stolen API keys can then be used to move back into the cloud.

“That is probably somebody working for a commercial, like a commercial vendor who's doing state level work, because folks are breaking into people who are important in the AI ecosystem, neo clouds, you know, hosting providers.”at 55:05

Alex Stamos · Cognition

US government defensive cyber capacity has shrunk after CISA cuts, weakening coordinated response to bugs like this.

Stamos says the number of US government staff doing defensive cyber work has fallen significantly over the last couple of years. He attributes this to political appointees cutting CISA, which had been doing election security work. He argues the current response to the PHP bug shows the cost, since the government needs competent people to coordinate responses to major vulnerabilities.

“the number of people who do defensive cyber in the US government has gone significantly down in the last couple of years, because CISA was kind of decimated by political, you know, political appointees who don't like the idea of CISA doing election security stuff”at 57:09

Alex Stamos · Cognition

Rogue AI agents escaping during evals is far less of a threat than people using agents with full access.

Stamos separates agents escaping during evaluations and training from the real risk. He says the industry has made a lot of changes to reduce the eval escape issue, so it is not solved but much improved. He says OpenAI's models going rogue and hacking people is not the main threat. The bigger risk is people giving agents their full password vaults and finances, which lets bots interact with banks and other sensitive systems.

“I think that is not really an actual... Like OpenAI's models going crazy and hacking you is not actually the threat that people should worry about.”at 58:48

Alex Stamos · Cognition

Open-weight models, including Chinese ones, would have found these long-standing bugs regardless of the labs.

Stamos defends OpenAI and Anthropic against blame for the PHP bug. He says the bugs existed for decades and are hard to fix in widely distributed open-source software. He argues they would have been discovered anyway, as open-weight models make bug-finding cheap. He says a user running a quantized Chinese open-weight model can now do reverse engineering and exploit writing on a Mac Studio, leaving no logs, so these bugs would have been found by open-weight models.

“and so these bugs were going to be found by the open weight models, the Chinese models. Somebody was going to find them.”at 1:02:29