Complex systems, wicked problems. Society, technology, science and more. @Princeton professor. @NYTimes columnist. My newsletter @insight https://xitter.cf/t.co/6Ky01N9JwA
floating in a most peculiar way
Joined August 2009
- Tweets12K
- Following738
- Followers38K
- Likes24.1K
Pinned Tweet
My Princeton AI and society team is at the @IASEAIorg AI summit in Paris.
My Neurips keynote for our approach:
slideslive.com/39055698/are-…
Our paper:
ai.sociology.princeton.edu
Say hi especially if you are *utterly* unimpressed by the current doom or hype models.
Wrong nightmares!
The first incident has ZERO to do with AI — the cert authority used 4 bit entropy instead of 384 🫠.
In the second, AI is heavily on the defense side of a long-existing problem finally being fixed.
AI creates issues but important to understand them right.
Patch your software.
Two classes extremely bad cyber vulnerabilities disclosed in the last 24 hours. The first potentially impacts everyone online. The second impacts nearly everyone running a server.
tl;dr: Everyone should upgrade their OS and Browser as soon as possible. If you run a server you should upgrade PHP (or ensure it’s removed if you’re not using it) as soon as possible.
The first class of vulnerabilities is a set of root TLS certificates that were discovered to have a trivial private key. Literally the number 4. These roots were controlled by #eMudhra. The problem is that with that an attacker could create a certificate and, if they could get in the traffic flow, intercept traffic to any site or service. The solution is to stop trusting certificates issued by these vulnerable roots. Upgrading your browser and OS is the fastest way for most Internet users to do this.
The second class of vulnerabilities doesn’t apply to most consumers directly, but does apply to most anyone running a server. These vulnerabilities have been dubbed #php2shell. PHP is one of the most widely used web programming languages. It’s estimated 70% of websites run on PHP. These vulnerabilities allow an attacker to send a crafted set of queries to a server running PHP and potentially take over the server. From there they could potentially read, modify, delete, or repurpose anything on the server. The solution is to upgrade PHP to the latest version ASAP. Even if you’re not using PHP for your web development, it’s often included by default. So make sure it is either upgraded or fully removed.
@Cloudflare received advanced warning of both of these classes of vulnerabilities. We’ve ensured that where we act as a proxy we are no longer trusting the vulnerable roots. We are also reviewing certificate transparency (CT) logs for any sign the vulnerability was exploited before its disclosure. We have not found any evidence but CT logs are not perfect and a sophisticated attacker who knew about the weakness could have exploited it without a new certificate being logged.
For the PHP vulnerabilities, we have deployed mitigations available for all customers for free. Because of the complexity of the vulnerabilities and number of ways it can be exploited, we also released new WAF capabilities that are unique in the industry to catch the exploit based on the response. These too are available for every Cloudflare customer for free.
Even if you are taking advantage of Cloudflare’s protection, it is important to upgrade your OS, browsers, and PHP installations as soon as possible. We can provide a strong bandaid, but the best protection is to fix the underlying problem.
If you’re interested in learning more, you can read our blogs:
Vulnerable roots: blog.cloudflare.com/emudhra-…
PHP vulnerabilities: blog.cloudflare.com/php-vuln…
“I should have rewritten everything in Rust years ago”. 😂😂
The funniest bit is yeah! That’s probably directionally correct for where this should go, if we actually want to do something! AI can make this easier! All we need is competent management of the transitio—
Never mind.
Above post but in substantive form.
This was bound to happen, AI is helping us fix things now, which is great, but the legacy code is not safe. Maybe we will finally contemplate moving forward.
Today's PHP apocalypse is akin to an overdue forest fire for the legacy World Wide Web, that long tail of at least 40M+ websites running PHP. It's the end of an era: the carnage kicking off now plays into a larger ongoing trend for the decline of the human-readable web over the coming years
Realistically, despite anyone's best efforts, some uncountably large number of sites are not going to be saved by fire brigades today. These sites are naked and vulnerable and being compromised as we speak. There is no happy ending, only mayhem.
This day was always coming. As long ago as 15 years ago, I was warning friends to pick another tech stack for their blogs and their websites, and I'm sure I wasn't alone in doing so. On my part, I felt that the quality of the PHP engine and runtime was a ticking time bomb, and that's only become more evident ever since
That bomb has gone off a number of times since, including as critical remote code execution incidents in 2012, 2019, 2024, and now. In total, PHP has amassed some 750+ CVE reports to its name and counting. And as might be expected, the pace of CVEs has picked up this year compared to recent years
This crisis is unlikely to be contained. You don't need Mythos access to find problems in the PHP runtime, just point Fable or Astra at it. What we are witnessing is growing strong selection pressure against large hairy legacy code bases in C/C++
Nothing better sells Rust's memory safety and ownership constraints than trying to do without them in a code base like PHP's--at some point even the clanker will begin suggesting that "you know, if I might just rewrite this wee little bit in Rust..."
Meanwhile, numerous personal blogs and other cobbler's-children kind of projects are, no doubt, in the process of getting a long overdue migration to static site generators. And another bunch is getting purged off the net altogether, as the easiest and most economical response to getting hacked. Thank goodness for the Wayback Machine!
zeynep tufekci retweeted
NEW: Trump’s mega Venezuelan oil deal, signed by Hegseth and Rubio, benefits an unknown group of investors and was intentionally crafted to keep the next U.S. president or future leaders of Venezuela from undoing it. @propublica saw the secret contract: propublica.org/article/trump…
zeynep tufekci retweeted
SCOOP: Julio Herrera Velutini, a Venezuelan billionaire, really wanted a pardon. He got it not long after his daughter hired Alina Habba, a lawyer close to Trump, in a deal worth more than $50 million. Habba was also working at DOJ at the time.
nytimes.com/2026/10/20/us/po…
zeynep tufekci retweeted
NEW @citizenlab report We uncovered an Israeli firm using online ads to run vast spy operation
This is the first-ever public confirmation of zero-click mercenary spyware delivered via targeted ads.
45 devices at @AJEnglish confirmed infected
citizenlab.ca/research/how-d…
zeynep tufekci retweeted
"So this is the phenomenon we’re dealing with: you have businessmen selling a technology, a Nobel Prize winner operating in the realm of science fiction, and this 26-year-old [Jacob Coxon] telling you: 'I’m on the inside, and I can see it.' Then, of course, you can say that journalists bear some responsibility, and that’s partly true, but it’s only natural: how can you not write an article about it when all the other journalists are doing so?"
This full interview of Michael Jordan is well worth reading. The other MJ occupies the same position in basketball roughly as this one in machine learning
di.ens.fr/~fbach/MJordan_ITW…
Here you go folks. Same old, same old. Nothing to do with AI. They connected to the Internet without patching for known vulnerabilities.
Remember, the FBI just lost sensitive medical files of thousands of agents because their contractor, Accenture, didn’t patch. Not AI either.
"A Shodan scan on Aug 19 still found 3726 of these controllers accessible from the public internet."
French supermarkets had their remote-controlled fridges wiped out with 50 C heat after they connected reportedly vulnerable Danfoss fridges to the public internet without patching for known vulnerabilities.
Victims include Intermarché, Coopérative U, Carrefour Market, Monoprix, Leclerc, and Super U, with French outlets reporting that the attack affected between 250 and 350 stores, causing significant financial losses and food waste.
infosecurity-magazine.com/ne…
zeynep tufekci retweeted
Anthropic and OpenAI have sought significant control over what outside evaluators can publish, concerned that findings could reveal trade secrets, according to two people familiar with the discussions. reuters.com/world/behind-ai-…
zeynep tufekci retweeted
See this is part of my complaint about the conversation right now, so many of the basic facts bulleted here are just plain wrong. But when we have people with a policy background and not necessarily a technical one, this becomes a larger problem because they cannot read the room or their own contradictions.
- The agents *were* sandboxed.
No, the sandboxes were flawed, the agents were not properly sandboxed. Anyone in cybersecurity can tell you this. Its either sandboxed correctly, or its not, it wasn't sandboxed. Call it a litter box if you want.
- The agents *were* given specific constraints on how to succeed at the benchmark, and flagrantly contravened them.
Giving agents an initial prompt then running a long event horizon job, the only thing you are actually going to expect without monitoring them, is they are going to drift from the original instructions as more context builds up, this is beyond basic analysis.
- The agents did *not* go hunt for the answers to the test (they already had the answers, by cheating); 1200 agents coordinated via tens of thousands of messages over several days on much more ambitious and weird attempts to *not be caught* for cheating, and hacking Hugging Face was one side quest along the way.
So you are saying the agents were sandboxed and fed previously unknown zero day vulnerabilities to exploit the sandbox without hunting for them? If not the agents were hunting for the answers. Found them, then covered it up. The contradictions here are wild.
This is why having technical rigor in these conversations is so important because otherwise we get contradictions in the same post. I don't like calling it out, but after spending a week with @MarcoFigueroa we discussed many of these issues in depth, and continuing to see such bad information coming out about it is quite disturbing.
You can have all the policy degrees in the world, without the technical rigor to understand what you are talking about, or the decency to consult the cybersecurity experts to update your assumptions, you are not going to be the right person to be disseminate this information for the general public... it only creates more noise when we need more clarity.
Unfortunate for someone playing a senior role in US AI policy (via the new SI Force) to get so many basic facts wrong here.
- The agents *were* sandboxed, and used zero days (ie previously unknown cyber flaws) to escape their sandboxes
- The agents *were* given specific constraints on how to succeed at the benchmark, and flagrantly contravened them
- The agents did *not* go hunt for the answers to the test (they already had the answers, by cheating); 1200 agents coordinated via tens of thousands of messages over several days on much more ambitious and weird attempts to *not be caught* for cheating, and hacking Hugging Face was one side quest along the way
I know there has been an avalanche of different incidents over the past few months, and in some cases the sandboxes were misconfigured or it's not clear how far the agents strayed from what they were told. I have sympathy for anyone struggling to keep up.
But the HF incident is distinct, and is actually extremely informative about where AI development is and where it's going. Fortunately there are public writeups that do let us understand at least the basics of what happened. I hope Shyam reads them soon.
zeynep tufekci retweeted
Many of Elon Musk's former DOGE foot soldiers have leveraged their government stints to fuel private-sector ambitions: building companies that tap government largesse or that deploy knowledge they garnered during their brief tenures in Washington. wapo.st/4ye0DsQ
Deleted the top because of Andrew’s point below (which I had already added to the thread but it’s distracting from the real issue) but adding here to reiterate: a meh analogy at worst but in no way “deny working cures to the dying”.
It was a reward hacking scenario, clearly.
It’s perhaps pointless in this environment to keep trying to explain but what reward hacking might look like in a medical environment isn’t hard to imagine.
See below thread.
