Posts

Home agents aren't starved for tokens, they're stuck in the queue

Image
Local inference has a bottleneck that nobody talks about, and it isn't tokens per second. It's one GPU. Run an agent that spawns subagents and suddenly five "parallel" workloads are all queued behind the same card, doing something very close to nothing while they wait for their turn. Nvidia's answer, announced at IFA 2026, is a free, open-source beta called PAIR (Personal AI Router) for Windows, Linux, and macOS. The name is misleading on purpose: it is not a hardware router. It's software that discovers compatible machines on your LAN (mDNS), pairs them with a six-digit code, and encrypts the channel with mTLS. Compatible means RTX 20-series and newer, RTX PRO, DGX Spark, and — worth repeating because it matters — Apple M4 and newer. The subtle-but-important part: PAIR doesn't teach your agents a new API. It proxies the Ollama and LM Studio endpoints you already run, and each model still runs entirely on one machine. It's workload distribution, not co...

Sality Is Down. The 15,000 Infected Machines Aren't.

Image
A 23-year-old botnet just got killed — and the killer wasn't a server seizure or a takedown order. Sality, a peer-to-peer malware network that started in 2003 as a classic file-infecting virus (the kind that latched onto executables and spread over network shares, USB drives, and file-sharing networks), has been infecting 15,000+ machines for two decades. International law enforcement, CrowdStrike's Counter Adversary Operations team, and the Shadowserver Foundation just cut its operator off from the entire fleet — by poisoning the one thing every Sality bot trusts: its own peer list. Every Sality bot keeps a list of "super peers," publicly reachable infected machines that form the P2P backbone, and it refreshes that list every 40 minutes. The operation exploited exactly that habit: researchers invalidated the legitimate super peers, swapped in defender-controlled sinkholes, and waited for infected machines to come online on their normal maintenance cycles so the remai...

CIP Just Became the Off Switch: Rockwell's Six-Advisory CISA Drop

Image
On September 1, CISA quietly dropped six ICS advisories, and every single one of them pointed at the same company: Rockwell Automation. The batch covers RSLinx Classic, the Logix controller platform (ControlLogix, CompactLogix, GuardLogix, and the 5480s), Factory Historian ME, the Redundancy Module Configuration Tool, and FactoryTalk Activation Manager. The headline isn't the count. It's the protocol. Across the whole pile, the common attack surface is CIP, the Common Industrial Protocol, and a single crafted CIP message is enough to take things down. In RSLinx Classic, four different flaws (all versions 4.50 and below, CVSS 8.6) let a malformed or oversized CIP packet crash the service, and in the Logix platform an improper length check on a CIP message pushes the controller into a Major Non-Recoverable Fault. No credentials, no user click, no exploit chain. A packet on the wire is the whole attack. Here's the part that should bother anyone who has ever stood in front of...

Debian Just Decided AI Code Is the Human's Problem

Image
Debian has finished the loudest AI argument in free software, and the winning policy is a masterclass in saying nothing with enormous force. The project's General Resolution on LLM usage closed on August 28 with a field that included an outright ban on AI-assisted contributions, and the ban died ugly: amending the Social Contract needed a three-to-one supermajority, and it finished at a ratio of 0.560. The winner, “Responsible Use of Generative AI,” took roughly 575 ballots from 425 unique voters, and it opens with the most Debian sentence possible: “Debian neither endorses nor prohibits the use of generative AI tools.” For a project that once had to vote on what its project lead was called, that non-endorsement is the strong position, and it's worth parsing because the part that says nothing is doing the heavy lifting. The part that has teeth is the responsibility clause. Every contribution, “regardless of how and with which tools they were produced,” must satisfy the same sta...

The CPU Bottleneck in Agents Just Became a Product

Image
There was a more physical way to announce a CPU than I expected. NVIDIA's vice president for hyperscale and HPC, Ian Buck, physically delivered AWS's first Vera CPU server in Seattle, the old-fashioned hand-off style, after similar deliveries to Oracle Cloud Infrastructure, Anthropic, OpenAI, and SpaceXAI. Vera is NVIDIA's first CPU purpose-built for AI agents: 88 custom Arm "Olympus" cores aimed at the unglamorous half of agentic work that never shows up in GPU benchmarks — the sandboxes, tool calls, orchestration layers, and long-context retrieval that surround every inference call. The timing on the announcement is the interesting bit. It shipped alongside news that AWS and NVIDIA are expanding their 16-year partnership with 2 million additional GPUs and Vera-based infrastructure landing inside AWS itself. A "built for agents" CPU walking into your cloud provider's datacenter is a different statement than a press release about one. Why should anyo...

2.9 Billion Commits a Month, and Nobody Asked Who Checks Them

Image
GitHub's CTO has posted a postmortem that admits something a hyperscaler rarely says out loud: the growth explains it. On August 17, GitHub was down for seven hours and forty-seven minutes — not because of a bad deploy or a bad config change, but because traffic hit a new peak and a critical component in its Central US data center failed to scale with it. Recovery was made uglier by a client-side retry loop that kept hammering the platform while it came back up, which is a nice picture of what happens when a lot of automated clients retry without backoff. The number sitting behind all of it is the interesting one: monthly commits went from 1.4 billion to 2.9 billion in four months. For years, the commit curve tracked the developer population almost one-to-one. It just stopped. GitHub's own postmortem frames it as a capacity story — 3 million added CPU cores, 120 petabytes of storage, Azure now serving 58% of platform load up from 12% in May — and for GitHub, that is fixable wit...

safeUnserialize Was Never Safe

Image
There's a special place in plugin development for function names that get things wrong in a way that becomes the whole story. GiveWP, the WordPress donation plugin with more than 100,000 active installs, shipped a helper called safeUnserialize() that, as it turns out, was neither safe nor the end of the attack chain. CVE-2026-82222 is a CVSS 10.0 unauthenticated remote code execution flaw, and per the Patchstack writeup, the winning move is a chain of three individually boring-sounding pieces: that broken "safe" unserializer, a donation-processing flow that happily stores attacker-controlled serialized objects in the site's database, and a gadget chain in the libraries GiveWP bundles that can turn those stored objects into arbitrary system commands. On versions 4.16.5.1 and earlier, all an attacker needed was a default install with one published donation form and one active payment gateway. The fix, version 4.16.7.2, landed last Thursday — after Patchstack rejected s...