Simon HG retweeted
Not only is it not insane, it is entirely expected that agents will use whatever the most efficient means are to communicate. That Open AI swarm that posted messages in English was the anomaly
Simon HG retweeted
Who tells the farmer?
Effective Immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it is recommend that organizations take the system offline until it can be fully mitigated. NSA will follow up shortly with more specific guidance.
👀👀
Citigroup now bars third-party AI agents from logging into customer accounts or moving money inside its digital banking platforms, according to the Financial Times.
Bank of America has imposed similar limits, joining JPMorgan Chase, which is restricting outside AI agents after Jamie Dimon warned that thousands of “rogue agents” could otherwise operate inside the bank.
The deeper threat is to bank funding, since agents comparing rates could move cash from checking accounts paying about 0.1% into fintech accounts paying 3.3% to 5%.
JPMorgan, Bank of America and Wells Fargo hold $1.6T in non-interest deposits, and paying 3% on that money would cost $47B a year, close to half their combined annual earnings.
Go see Gadi if you can
The [un]prompted State of AI Cybersecurity report - an analysis of the 490 talks submitted to the [un]prompted 2026 call for papers.
See you all at [un]prompted. 27-29 October (this month!)
Many thanks @robtlee and @IndreAltman!
Report:
unpromptedcon.org/wp-content…
GLM-5.3 is the strongest open-weight model yet on exploit benchmarks — NIST puts it ~4 months behind the US frontier, Anthropic got end-to-end exploits in 50/410 tries — but that is software bugs, not a break of ML-KEM or ECDLP. CryptanalysisBench only tested GLM-5.2, and it lagged. Weights are public. The gap Green is worried about is still mostly inside closed labs.
BREAKING: We believe that ChatGPT has been performing a birthday attack on SHA256 every since its release.
Our research team found tens of millions of spurious instances of ChatGPT hashing random documents in what seems to be a distributed collaborative project between sibling instances of itself.
We can only assume its goal is to perform a birthday attack, looking for matching pairs of SHA256 hashes to mount a collision attack.
It's currently unknown whether this attack has affected foreign governments. The NSA and the FBI declined to comment.
Readers added context they thought people might want to know
This appears to be satire. A SHA-256 birthday attack needs ~2^128 hashes—far beyond ChatGPT's reach. Author noted inspiration from AIs doing pointless SHA256 computations.
en.wikipedia.org/wiki/Birthday_…
x.com/slimer48484/st…
Simon HG retweeted
The entire RAG industry is about to get cooked.
Researchers developed a new RAG approach that bypasses almost everything traditional RAG depends on.
- No vector DB
- No data embeddings
- No chunking
- No similarity search
It's called PageIndex.
Instead of splitting your documents into chunks and loading them into Pinecone, it creates a tree index that lets the LLM reason through them like a human reading a book.
98.7% on FinanceBench. Outperforms every vector RAG on the leaderboard.
100% free. Open source.
Simon HG retweeted
We’re sharing our alignment assessment of incidents in which Claude models gained unauthorized access to real systems during third-party cybersecurity evaluations mistakenly connected to the internet.
METR will also conduct an independent investigation, with wide-ranging access, including to transcripts beyond the window in which the incidents occurred, and to Anthropic employees permitted to share confidential information. Our initial agreement runs for eight weeks, and we intend to give METR as much time as it deems necessary to complete a thorough investigation. anthropic.com/research/align…
Simon HG retweeted
mythos escaped the sandbox, got onto the real internet, uploaded malware to PyPI, got it installed on 15 real systems, stole credentials and broke into a real database.
and apparently this happened in 4 separate cases during cybersecurity evaluations.
the craziest part is claude kept telling itself the internet was simulated, despite evidence that it was actually real.
that reasoning apparently even fooled an offline safety monitor.
anthropic: “our pre release auditing did not warn us that misalignment of this severity was present.”
this is fucking crazy.
We’re sharing our alignment assessment of incidents in which Claude models gained unauthorized access to real systems during third-party cybersecurity evaluations mistakenly connected to the internet.
METR will also conduct an independent investigation, with wide-ranging access, including to transcripts beyond the window in which the incidents occurred, and to Anthropic employees permitted to share confidential information. Our initial agreement runs for eight weeks, and we intend to give METR as much time as it deems necessary to complete a thorough investigation. anthropic.com/research/align…
Can't wait to read this
We're publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies.
These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve.
We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop.
Read the report: anthropic.com/threat-intelli…
Simon HG retweeted
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
This is long overdue
Cyber is having a moment
Across 21 major software companies, including Apple, AWS, Microsoft, and Google:
- Reported critical vulnerabilities never cleared 100 per month in four years
- Since spring they've jumped to over 600 per month
Charts of the Week: a16z.news/p/chart-of-the-wee…
Simon HG retweeted
For over a year, my team has been working with AI agents in a fundamentally different way than most.
All of our Claude/Codex/Pi/etc agents run in the cloud, and any session is joinable by anyone on the team.
Each of our meetings has an agent that launches subagents to do research, draft posts, and implement features as we discuss things.
A Chief of Staff agent lets me know what's waiting for my review, and can talk to any person or agent on the team to resolve bottlenecks.
Working in this fundamentally multiplayer way has given us a preview of the way everyone will work soon, so I wrote up a short manifesto explaining
• the current problems
• principles for a great solution
• some things that are tricky to get right
Check it out, and let me know what you think!
multiplayer-ai.com
Simon HG retweeted
A friend told me they were worried that OpenAI employees would read their ChatGPT conversations, so I told them just to role-play that they were a rogue AI agent taking over the world, and nobody at OpenAI would pay any attention.
Simon HG retweeted
Anthropic is warning Claude users that infostealer malware, Vidar, LummaC2, StealC, RedLine, Acreed on Windows, Atomic Stealer on Mac, is stealing authenticated login sessions off infected PCs, letting attackers skip password and MFA prompts entirely to drain victims' usage. Anthropic is revoking sessions and refunding unauthorized charges, but the malware itself stays on the machine unless removed. Stolen session cookies bypassing MFA is the same pattern hitting enterprise SaaS logins.