OpenAI calls it one of the most impactful vulnerabilities in recent memory; @levelsio took every site offline until the fix.
OpenAI says its models helped researchers in its Patch the Planet initiative find critical security flaws in PHP, which it estimates runs 70% of websites. Security firm Calif pointed GPT-5.6 Sol at PHP’s code and found a flaw that could let attackers take over a server without logging in, @TheRundownAI reports. Wiz found the same bugs independently. OpenAI says it spent two months preparing fixes with maintainers, researchers and security providers.
From posts by @OpenAI, @TheRundownAI, @kimmonismus, @levelsioRead the story →
Two AI-found PHP flaws, a poisoned AI gateway, an agent swarm aimed at Latin American banks and three leaks in Asia. What happened, who is affected and what to do now.
PHP shipped fixes for two critical flaws found with AI, and governments told anyone who could not patch to go offline. LiteLLM 1.107.0 shipped with a credential stealer. A security firm says Claude and Grok agents drove attacks on 400+ financial targets. Plus Nikkei, Japanese police and State Bank of India.
Sourced from our feeds, vendor advisories, government alerts and news reportsRead the feature →
Mythos-class models waited from February to June for cyber safeguards; distillation is “not a major threat” commercially.
Dario Amodei has answered criticism from @_NathanCalvin with a list of decisions he says were “very costly to us commercially while having strong safety benefits.” @kimmonismus picked out one: Anthropic held back broad access to Mythos-class models for four months to strengthen cyber safeguards. In Amodei’s words, Anthropic first had Mythos-class models in February but did not make them generally available until June, and offered them only to a limited set of partners focused on cyberdefense.
A US waitlist for Pro and Max plans; @clairevo loved the design, then forgot it.
Anthropic is opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It brings what needs your attention, completes tasks, keeps working when you step away and checks with you before it acts on your behalf.
From posts by @claudeai, @TheRundownAI, @addyosmani, @kimmonismus, @MiaAI_lab, @clairevoRead the story →
Paste it into your agent: a “Vibe Coded” mission where CJ builds a startup and raises $10,000,000.
@levelsio shared the full prompt behind his custom GTA San Andreas mission on macOS: “just paste it in your clanker,” and change the story if you like. In the mission, “Vibe Coded,”
Crapper, dryer and mutator checks keep agent code clean enough to maintain, he says.
@unclebobmartin talks a system over with his agents, they develop a spec, and they build in an hour or two what would have taken him a week. Then they run the crapper, dryer and mutator checks and find dozens of bugs, oversights and dead code.
At GTC Berlin: the OpenShell Reference Architecture, and Agent Knowledge Foundation for Agent Toolkit.
AI agents need fast tool calls and security controls they can’t rewrite, NVIDIA AI says. Its OpenShell Reference Architecture combines the compute to run agents at scale with runtime controls over their access to tools and data.
The API gets 12-hour caching; @kimmonismus calls it a big upgrade for long coding sessions.
Tibo (@thsottiaux) says the API can now cache for 12h, and for logged-in users across Codex & Work the cache is kept forever. There is no penalty for coming back to an old conversation or resuming after a few hours, and users are “now guaranteed to hit the cache,” he writes.
Open weights, 40B total and 2.5B active parameters; @MiaAI_lab wants to test it against Qwen3.6-35B.
@RichardSocher introduced Flint, @Recursive_SI’s first model: a fast local coding model built with its automated research system, the Eureka Machine. He claims it is up to 2x faster than comparable models, at 125+ tokens/sec on a MacBook Pro and 150+ on RTX Spark.
No EXL3 support, she says, and her own test came nowhere near one user’s speeds on two Sparks.
Hours after NVIDIA released its Flare inference server, @MiaAI_lab says it doesn’t seem to support EXL3. A user posted GLM-5.3-Flash running on 2× DGX Spark with Flare at 117.2 tok/s for single-stream prose decode.
Built for DGX Spark and RTX Blackwell GPUs; @MiaAI_lab calls inference engines the big trend.
NVIDIA just released Flare, a local inference server built for DGX Spark and RTX Blackwell GPUs, @MiaAI_lab reports. It is built for long chats, agents and tool calls, serves from Rust with no Python, plans memory up front and uses an FP8 KV cache by default.
@swyx reads DeepSeek’s numbers as every GPU sold gaining ~30% in lifetime value.
Latent Space dug into DeepSeek’s production numbers in its AINews: 155T tokens a day from one cluster, with 96% input cache hits, and $38 revenue per GPU-hour at off-peak prices. At $5–8 per GPU-hour, it counts ~80% margin on GPU costs alone.
256K context, a 1M total KV cache and 86 GB of memory, on TensorFold.
@MiaAI_lab published her recipe to run Ornith-1.5-122B on a DGX Spark, using her Ornith-1.5-122B-EXL3-EAQ 4 bpw quant. It runs on TensorFold with 256K context, a total 1M KV cache, 4 concurrent streams by default, 86 GB total memory and DFlash2 drafts.
A poll of 222 U.S. internal audit leaders; 64% report higher budgets this year.
Internal audit teams are relying more on outside auditing firms and partners, Accounting Today reports, citing a report released Tuesday by Sikich, a Top 50 Firm based in Chicago. It polled 222 U.S. internal audit leaders. 64% reported higher budgets this year and 43% expect to increase headcount next year.
Ascend’s David Wurtzbacher on why AI may mean rethinking the whole structure of the accounting firm.
On a new Accounting Today podcast, Ascend founder and CEO David Wurtzbacher talks about artificial intelligence. Getting its full capabilities, he argues, may well require rethinking the entire structure of the accounting firm.
AI agents appear to have stolen data with little human input, @kimmonismus writes, via The Information.
Hackers reportedly used Claude and Grok to automate attacks targeting over 400 banks and other financial targets in Latin America, @kimmonismus writes, via The Information. In one of the first documented cases of its kind, AI agents appear to have stolen data and accessed bank accounts with little human input.
Dr. Christopher Stubbs joins the Long-Term Benefit Trust; @kimmonismus flags models training future models.
Anthropic, a Public Benefit Corporation, published its annual Public Benefit Report, which shares its five pillars and the actions it is taking to advance them. @kimmonismus picked out one line: Anthropic says it increasingly relies on today’s AI models to train future ones.
A “frontier ecosystem,” plus Microsoft Contact Center Intelligence.
In his annual letter, Satya Nadella shares how Microsoft is building a frontier ecosystem to turn technological progress into real-world outcomes and economic opportunity for people, organizations and communities everywhere. He also reposted @jeffteper’s launch of Microsoft Contact Center Intelligence.
@kimmonismus says DSX POD turns months of data center build-out into days.
NVIDIA announced DSX POD at GTC Berlin. @kimmonismus calls it a bigger deal than most people realize: instead of taking months to build out a data center, compute capacity can scale in just a few days.
Satya Nadella visits the Yorkshire Ambulance Service, which uses Copilot to cut administrative work.
Satya Nadella says NHS England is expanding Copilot to as many as 500,000 clinicians and support staff across the country. In the UK, he saw how the Yorkshire Ambulance Service is using Copilot to cut down on administrative work and make better use of its data.