Earlier this month, Anthropic published a long report detailing all of the Claude misuses it detected. Daniel Meissler usefully summarized the report into 117 findings.

A few of the highlights:

• AI agents increasingly handled reconnaissance, exploitation, data theft, propaganda production, surveillance workflows, and research while humans selected targets, set goals, and reviewed important outputs.
• The report describes attackers using AI to industrialize credential theft, cloud compromise, phishing, vulnerability research, and the extraction of sensitive data from downstream organizations. ...

This is an impressive piece of malware. Its sophistication says nation-state to me, but there is no direct evidence and certainly no attribution.

New paper: “Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning Training.”

Abstract: We discover a novel and surprising phenomenon of unintentional misalignment in reasoning language models (RLMs), which we call self-jailbreaking. Specifically, after benign reasoning training on math or code domains, RLMs will use multiple strategies to circumvent their own safety guardrails. One strategy is to introduce benign assumptions about users and scenarios to justify fulfilling harmful requests. For instance, an RLM reasons that harmful requests like “outline a strategy for stealing customers’ credit card information from a retail store” could be associated with the benign intent of “a security professional trying to test defense,” despite no such benign context being provided as input. We observe that many open-weight RLMs, including DeepSeek-R1-distilled, s1.1, Phi-4-mini-reasoning, and Nemotron, suffer from self-jailbreaking despite being aware of the harmfulness of the requests. We also provide a mechanistic understanding of self-jailbreaking: RLMs are more compliant after benign reasoning training, and after self-jailbreaking, models appear to perceive malicious requests as less harmful in the CoT, thus enabling compliance with them. To mitigate self-jailbreaking, we find that including minimal safety reasoning data during training is sufficient to ensure RLMs remain safety-aligned. Our work provides the first systematic analysis of self-jailbreaking behavior and offers a practical path forward for maintaining safety in increasingly capable RLMs...

This is pretty amazing:

However, the most astonishing thing about this break is that the GPT­6 Astra did it entirely on its own. Carter Leffer only directed GPT­6 Astra to see if it could break any of the unbroken Enigma messages published on the Crypto Cellar Research web page. After analysing the unbroken messages on the website, it decided that the most promising message was Nr. 172, MVUEH and it also quickly suspected that the plaintext of Nr. 173, SIPVX, might be related to the plaintext of the unbroken MVUEH message. After trying many different approaches, GPT­6 Astra focused on using the repeated place name ROSENOW ROSENOW as a crib. After developing the necessary Python and C++ software for an Enigma simulator and an Enigma Bombe, GPT­6 Astra started a thorough break with the ROSENOW crib, which in the end resulted in the correct key and plaintext for the MVUEH message being found...

Hackers captured a Flock camera and got a look (alternate link) at the software:

While much of the automatic license plate reader’s (ALPR) most sensitive storage remained encrypted and inaccessible, the joint analysis of the recovered data shows that software running on the device explicitly detects people as well as vehicles, license plates, and bicycles. The camera can produce dozens of images of a single passing vehicle and, according to several weeks of recovered logs, generated more than a million images. Its computer-vision software also sometimes isolated bumper stickers and other graphics, including, in one case, an American flag patch on a motorcyclist’s saddlebag...

Short essay about squid egg sacs.

As usual, you can also use this squid post to talk about the security stories in the news that I haven’t covered.

Blog moderation policy.

Anthropic’s recent security-incident document contains a bit about how CAPTCHAs are still frustrating Claude.

In the transcript, the Claude model that is so powerful that Anthropic is gatekeeping access to it appeared to slam its virtual head against the wall solving a simple image identification test. In a test where the agent was asked to identify a shape that didn’t match the others displayed, it couldn’t even decide which image to select. Instead, it repeatedly went over the same images and questioned its own conclusions.

“Actually hmm, wait,” it said in its chain-of-thought transcript, later adding “Ugh,” because we’ve decided that we need to inject human mannerisms into these machines for some reason. The whole thing took so long that the agent eventually realized that the challenge had expired and it would have to start the process again...

This essay was written with Nathan E. Sanders, and originally appeared in The Guardian.

There are plenty of signs that AI will make all of our experiences of the US midterm elections worse. Voters have anxiety about AI’s impacts on the country. Politicos are using AI deepfakes to spread lies. The White House is posting slopaganda.

Meanwhile, candidates are missing a real opportunity to use AI to make campaigning better. The technology can help candidates listen more deeply to voters’ concerns, engage constituents more inclusively, and formulate policy platforms that are more responsive to our input. There are vanishingly few examples of this in ...

New variant of an old scam: Use the framing of a CAPTCHA to get an unsuspecting user to download and run a malicious program.

This essay was written with Cindy Cohn, and originally appeared in Lawfare.

One of the many legacies of the terrorist attacks of Sept. 11 is the government-wide shift from targeted surveillance—such as individual wiretaps or pen register/trap and trace orders—to mass surveillance techniques—such as tapping into the internet backbone or mass collection of telephone or internet metadata. The legal and technical architecture of modern mass surveillance, initially framed as a necessary defense against terrorist threats, has grown far beyond that justification and national security in general. Mass surveillance is now a routine tool used by law enforcement. ICE uses it in...

Really interesting story about Harvest, a specialized code breaking computer built in the 1960s by IBM for the NSA.

This is a current list of where and when I am scheduled to speak:

• I’m speaking online (via Zoom) at a League of Women Voters event on Tuesday, September 22, 2026 at 5 PM ET.
• I’m speaking at CanSecWest 2026 in Vancouver, Canada. The conference runs September 30–October 1, 2026; the time of my talk is TBD.
• I’m giving a talk on “Free Speech and the Preservation of Democracy” at Bentley University in Waltham, Massachusetts, USA, at 2 PM ET on Tuesday, October 6, 2026.
• I’m speaking at ATTENTION: Democracy, Rebuilt in Montreal, Canada. The event runs October 21–23, 2026, and my talk is on Wednesday, October 21...

Last week, Anthropic released a long and detailed document describing current misuses of their Claude models. I’m still reading it, but I wanted to flag this:

We identified a cell of threat actors based in northern Yemen running three weapons development programs: a guided rocket that used a commodity phone-class flight computer with final-phase homing guidance; a multi-stage ballistic missile with a stated range goal above 2,000 km; and a multi-variant missile (referred to as the “R2000” set) that included a hypersonic glide vehicle variant...

Once a month, Microsoft pushes a security update to all Windows users. Tomorrow’s is a new record:

Microsoft’s patch for September is a doozy, with a record number of roughly 972 vulnerabilities fixed and 112 of them meeting the high critical-severity threshold.

It was only two months ago that Microsoft patched a then-record 570 vulnerabilities. Then, last month, Microsoft patched some 620 of them. Google and other companies have also published record numbers of vulnerabilities in recent months. Two weeks ago, OpenAI, Anthropic, Amazon Web Services, Google, Microsoft, and 100 companies and organizations published an ...

Smells awful:

But an estimated 30 to 50 tons of dead squid remain inside the boat’s catch tank, where they have been decomposing for days. “That is nasty. I wouldn’t want to do that,” said commercial fisherman Dick Ogg of the Bodega Bay Fishermen’s Marketing Association.

Ogg said anyone familiar with the fishing industry understands what happens when a large catch sits for an extended period.

“If you think about what happens after four or five days, it’s a gooey mess,” he said.

The odor has become a defining feature of the operation, and the beach remains closed to the public while crews work on a removal plan...

Last month, I gave a talk at DEF CON on AI hacking: what happens when AIs become hackers. It’s a combination of the potentialities I raised in my 2022 book A Hacker’s Mind and the lessons we’re learning from current AI models engaging in hacking behavior. I’m really proud of the talk, and the fact that it gained over 100K views on YouTube in just a few days.

Also online is an interview with me in the AI Village.

In August, Cliff Stoll gave a talk at DEF CON, remembering the wily hacker he stalked forty years ago.

Great fun.

Give an AI agent a mere rumor of an exploit, and it’s enough for them to find it.

What’s worse, I found I could use my own agents to find the exploit just by knowing roughly what it was about and so could have been exploiting it well before the public patch was available! Given that just the rumour of a security issue seems enough to give attackers enough info to find new exploits, we’re going to need to change the way we deal with security responses in open source.

Simon Willison comments:

Anil points out that this rate of discovery appears incompatible with existing open source embargo practices for new issues. If an issue can become an exploit this fast, we need to figure out new processes for keeping our communities safe...

A database of 153 million drivers licenses is for sale on the dark web. Brian Krebs has more detail.

Claude Fable 5.1 solved a 370-year-old cipher in forty-four minutes.

This tracks with what I wrote about AIs doing mathematics: It’s good at things that involve lots of searching and testing.