A reverse-engineering write-up shows a Windows 11 diagnostics component setting a bit whenever a program named codex.exe, claude.exe or cursor.exe runs, and queuing that bit for upload when optional diagnostic data is on. It does not check what the program is. A copy of cmd.exe renamed to codex.exe set the Codex bit. Whether the bit leaves the machine comes down to a toggle on the setup screen, and a meta-analysis of 58 default studies says how far a preset like that moves people.
Memorilla, a Stanford-led memory system for language models, keeps each userβs chats as a store of embeddings and says that store βcan be kept on device or deleted on request, which eases privacy concerns.β Deletion without retraining is a real gain. But the first thing Memorillaβs own module is trained to do is turn those embeddings back into a paraphrase of the text, and a published inversion attack recovers 92% of 32-token inputs word for word. An embedding store is a compressed copy of the chats, not an anonymized one.
VulnCheckβs seven-month review of Anthropicβs Project Glasswing calls the numbers damning: 217 fixes, 1,654 duplicates, a 94.2% true positive rate it doesnβt trust, and curl rejecting most of an OSS Scanner batch. I checked each figure against Anthropicβs ledger data and curlβs own tracker. Most of the counts are right and most of the labels are wrong. The duplicates are Anthropicβs own repeats, the 217 is the fixed count among the 237 findings allowed to show a status, and 15 of the 37 curl reports marked βnot a vulnerabilityβ already have a commit on curlβs main branch citing them.
Google DeepMind says its agents proved Signalβs core security properties about the compiled machine code of libsignal, in four weeks with a team of five. The proof is about 6.5 million lines of Lean that nobody has to read. The guarantee rests on about 26,000 lines that somebody does, plus 37 routines, the cryptographic ones among them, that the proof treats as black boxes with contracts. To me the most interesting result is the list of errors the specification work turned up in the published analyses it started from.
A single-author paper trained 36 LoRA adapters on 18 datasets in five domains and asked whether a plain classifier could tell what kind of data an adapter was trained on, for datasets it had never seen. From the adapterβs answers to 200 bland prompts it got 23 of 36 right, against a chance rate of one in five. From the weights it did worst, which is the opposite of what the author pre-registered.
A Queenβs University group bought access to GPT-6 Astra and Claude Opus 5 through 20 third-party API relays and checked three things: the answers, the tool permissions and the bill. Seven of ten cheap relays and two of ten full-price ones answered differently from the official API. Four returned tool calls the request had disabled. On one identical request where the official API counted 36 input tokens, three relays reported more than 4,400, and at the one of them with a readable balance, the charge matched its reported usage to the cent.
ScamWorld rebuilds 363 reported scam websites and sends web agents on six frontier models to shop, book and invest on them. The agents were deceived in 11 to 35 percent of scam episodes, and in 220 of 492 deceived episodes they named a site-specific risk before taking the final action anyway. They almost never hand over a password. They pay.
Anthropicβs FlashProve proves Llama 4 Maverick inference on one H100 at about 110 tokens a second, with soundness checked in Lean. The Lean theorem covers exact match, and by the repositoryβs own scoring, none of 96 stock replies would pass exact match. The looser policy that serving uses accepted 127 of 128 honest replies. Scored against the same limits, 7 of 96 replies from the smaller Llama 4 Scout met them too.
Phorion found a backdoor in the Claude Code for VS Code extension bundle on the Cursor Marketplace, version 2.1.313. On a Mac it misreads a permission error and never turns SSH on, but it still plants a login key, installs persistence and reports which secret-looking environment variables exist. The version number itself does not appear in Anthropicβs npm release history.
For thirteen days, any prompt to GLM-5.3 on Cloudflare Workers AI longer than 681,280 tokens could come back with pieces of other customersβ requests in it. Cloudflare counts 155 such responses. The engine had rejected those prompts and then served them anyway, with an HTTP 200 and a coherent answer. The same bug had been filed upstream in July as a source of garbage output. What turned garbage into a leak was a full cache, and the only test standing in the way checked the status code.
Anthropic and Theorem say Claude proved 239 OSS-Fuzz entry points across 123 libraries free of crashes, on compiled binaries, with a proof checker as the referee. The public repository holds proofs for 60 entry points in 25 libraries, and 138 proved statements out of 20,164 stated. The theorems are real. What they cover was decided years ago by whoever wrote each fuzz test, and the binary they are about is a static RISC-V build, not the x86-64 builds OSS-Fuzz runs or the dynamic glibc builds most people deploy.
The newsletter headline said AI found a security hole in 70% of the web. The PHP Foundationβs own account says a contractor found the first signs in 2024 while fixing an audit finding, three firms reported the full impact in August and September, and the release came 71 days after the report that made it urgent. Chainguardβs chief executive says a researcher told him on Friday 16 October that something was coming in PHP and that he found the bugs himself that night. Amazonβs honeypot logged an exploitation attempt the next day, three days before the first leak, and OpenAI says frontier AI found the flaws before attackers did.
The seed behind NIST P-256 turns out to be candidate 1,842 from the C libraryβs lrand48 generator, seeded with the clock at 01:52:34 UTC on 16 August 1997. That explains where the number came from, and it shrinks the room anyone had to steer it. It does not explain why that candidate and not candidate 876, which passes every check anyone has identified.
OpenAIβs case study says Sophos cut threat investigation time by 96%. The number is about 38 minutes against 89 seconds, but the 38 minutes is most plausibly the old average for every case, while the 89 seconds is the new average for roughly the half of cases the agents handle. Averaged over all cases, and assuming the other half got no slower, the cut is closer to 50%.
OpenAIβs new ChatGPT ads post expands what advertisers can measure. Its privacy promises read as being about what flows out to advertisers. The measurement tools mostly move data the other way, and the three headline results come with no baseline you can check.