A top Cybersecurity and Infrastructure Security Agency official has called a security vendor’s decision to break the embargo on the recently disclosed PHP vulnerabilities “dangerous” and “inappropriate,” as cyber professionals grapple with how artificial intelligence is accelerating vulnerability timelines.
The company in question, Chainguard, released patches for the PHP bugs more than six hours ahead of the planned, coordinated release last week, according to the PHP Foundation.
Under coordinated disclosure processes, security firms like Chainguard receive vulnerability information in advance so they can develop patches and their users can immediately update their systems upon disclosure.
Chainguard’s decision has been criticized for putting others at risk, allowing hackers to study the PHP vulnerabilities before other organizations were ready to release patches.<![CDATA[]]>
“Any attacker could now study the fix in binary form while every other PHP user was still waiting for the release,” the foundation wrote in a blog post last week. “This disregarded weeks of work by many parties toward a coordinated disclosure.”
Chainguard CEO Dan Lorenc defended the decision in a LinkedIn post last week, writing that it was made “based on how easily we were able to independently discover the vulnerability and build an exploit, how much information was already out there on the coming vulnerabilities and fixes, and the evidence shared on active exploits in the wild.”
The PHP Foundation, on the other hand, said that despite some information about the vulnerabilities leaking in the two days prior to the targeted release date, its security team “did not identify any functional exploits at the time of release/public disclosure.”
Lorenc’s defense has been disputed and debated in online forums. And this week, CISA’s acting associate director for vulnerability management, Jay Gazlay, said publicly that Chainguard’s decision to break the embargo was “much to all of our danger.”
“I will say that that type of behavior is dangerous, and we’ve got to be careful about holding embargoes,” Gazlay said during an Oct. 28 conference hosted by Red Hat in Washington, DC. “We’ve got to be careful about holding the trust groups because the minute we stop behaving in in a manner that is honorable and trusting in this world, it gets ridiculous.”
“Breaking embargoes, that’s a choice that we continue to make,” Gazlay continued. “Anytime somebody behaves in a way that is inappropriate, if we’re like, ‘Oh yeah, it’s fine,’ they’re going to do it again. It is incumbent upon all of us within the ecosystem, when we see people breaking embargoes, to say that is not okay, because if we say anything other than that is not okay, they think that is okay, and they keep doing it.”
CISA declined to make Gazlay available for a follow-up interview on his comments.<![CDATA[]]>
Chainguard responds
In an interview with Federal News Network, Lorenc said he stands by his LinkedIn post last week, while welcoming the critiques of his company’s decision to break the embargo. Lorenc also said he respects Gazlay and agrees with the need to continue coordinating on disclosure.
“Coordinated vulnerability disclosure works on trust, works on understanding what the criteria is here, and I think that’s something we all have to get crisper on and do a better job on next time,” Lorenc said. “There’s a lot of this coming. I think a big one for the industry in general is going to be how do we move faster and get through these quicker.”
Lorenc has said Chainguard saw enough evidence that the PHP bugs were being exploited “in the wild” prior to the patch embargo to justify releasing fixes early.
Many critics of the Chainguard decision argue there wasn’t enough underlying evidence that the PHP bugs were being exploited by threat actors to nullify the agreed upon embargo on Oct. 21. Some research identified attempted exploits dating back to Oct. 17. But it took several days after the official disclosure for threat researchers to confirm a working exploit of the PHP bugs. CISA added the vulnerability to its Known Exploited Vulnerabilities catalog on Oct. 26.
Lorenc argued that there doesn’t necessarily have to be ironclad evidence of successful exploitation to justify releasing patches early and that such judgement-based decision making is “messy.”
“It has to be in someone’s hands that is trying to do bad things,” Lorenc said. “They don’t have to be successful.”
He said that the pace of vulnerability discovery and exploitation introduced by frontier AI labs is adding “a lot of friction and pressure” across software security supply chains.
“I think the real challenge everyone’s going to have to deal with is what happens when you’re sitting on one and can’t get it up-streamed in time, and then harm happens during that period,” Lorenc said. “That’s something I’d love to see CISA and other folks get involved in and start to work out a framework on. The PHP situation was a happy path in many ways. There are much rougher ones that I know everyone sitting on these vulnerabilities is dealing with today.”
The debate amid AI acceleration
Joe Slowik, director of threat research and cyber engineering at Dataminr, said while there may be compelling reasons to break security embargoes, he hasn’t seen any public evidence to back up Chainguard’s decision.<![CDATA[]]>
“The evidence since disclosure has proven there hasn’t been much of a critical concern because it took about a week before it appears this was effectively weaponized and it still hasn’t been weaponized at scale,” Slowik told Federal News Network.
Choosing to release the vulnerabilities early, however, gives attackers across the world more time to study the information and develop exploits, Slowik said.
“In the age of AI, that analysis cycle becomes even more compressed,” he said. “As soon as these items are disclosed, people now have a solid lead to figure out what changed. So even if there was or there were signs of some testing or experimentation around PHP, it basically opens the floodgates.”
But Slowik disagreed with any calls to freeze Chainguard out of future vulnerability disclosure conversations.
“This was a judgment call that we should learn from, and all organizations should internalize for, ‘Hey, maybe don’t do this for reasons that we learned about in this specific activity,’” he said. “As far as what that means for the future, though, it’s really unsettled right now. There’s a lot to be said for the way in which vulnerability disclosure and weaponization has kicked up in terms of speed.”
According to Politico, PHP Foundation Executive Director Elizabeth Barron said that while the foundation was “disappointed” with Chainguard, it hopes to “rebuild trust with them in the future to ensure we can continue to collaborate effectively.”
Meanwhile, Ramy Rahman, senior principal solutions engineer at ArmorCode, said the incident highlights how AI is helping both defenders and attackers. AI models helped discover the PHP bugs and develop patches, while it was also used to try and develop exploits.
“There needs to be a balance when it comes to responsible disclosure, but it’s tricky,” Rahman said.
AI is also muddying the waters around how vulnerabilities can be exploited, creating further confusion. Security researchers have reported seeing many cases of AI-hallucinated exploits associated with the PHP bugs on GitHub.
Gazlay described CISA’s vulnerability management team encountering similar issues.
“To exploit this PHP vulnerability, we had some of the best analysts in the world look at it for multiple, multiple days using great models, and we were able to generate a working proof of concept inside our own infrastructure,” Gazlay said during the Red Hat conference. “We then saw a lot of evidence come in that, when we say evidence … it was terrible. It was just slop, and I believe in vibe coding, but I also believe in somebody looking at the work.”
Slowik said the lesson from the Chainguard decision is “be a little bit more cautious, even if we think we’re under the AI gun, in getting these things pushed as soon as possible.”
“This is all hard, and we’re all trying to figure out how this new landscape, in terms of velocity and having lots of entities, whether it’s your frontier AI models or people using open-weight, openly available models outside of those ecosystems to research these things,” he said. “It’s a very unsettled space, and we’re learning a lot as we go along here.”
Rahman argued for “being empathetic” toward organizations involved in addressing software vulnerability management issues.
“We all have a vested interest to come together,” Rahman said. “This is one of those that we can really learn from each other here to maybe just put our guard down, maybe start to collaborate a little bit better, and also just have a better moral code of saying, ‘Hey, we feel like it’s important to protect the enterprise and get our fixes out there fast, and we also really want to enable our customers to be able to make their decisions on their own, so that we have more open information, people can make better decisions.’ And I think there was a portion of that that failed in this, and we could learn a lot from it.”
Copyright
© 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.
——————————————————-
Click Here For The Original Source.
