Systems

Attestation Proves a Hash. It Does Not Prove What the Hash Is.

Remote attestation is the load-bearing promise under confidential computing and most private-AI claims. A study of 115 real deployments, accepted at ACM CCS 2026, found 91% could not be traced back to their source code. One developer out of twelve said reproducibility was a priority.

The KEV Catalog Stopped Being a Patch List. Nobody Told the Dashboards.

CISA replaced the flat everything-in-KEV-same-deadline model with a four-variable risk matrix and started stamping machine-readable decision points onto individual CVEs. The actively exploited Chromium bug is the first clean worked example, and its clock runs out on 18 September.

10 min read

What Is a Sandbox Escape?

Arbitrary code execution inside a sandbox and a sandbox escape are two different things, and headlines routinely merge them. Here is what a sandbox is, why one bug is usually not enough, and how to check a claim yourself using the fields in the CVE record.

110 Million Installs a Week. The Funding Model Still Did Not Work.

Tailwind Labs is joining Shopify. The CSS framework stays MIT, but the paid products that funded it are closing to new customers. The last well-known example of fund-the-commons-by-selling-something-next-to-it just resolved into an acquisition, which means every option has now been tried.

Two Certificate Authority Keys, One Desktop PC, Sixty-One Hours

Somebody pulled the 512-bit root certificates out of Netscape 4.51, pointed a gaming CPU at them, and factored both. A third fell to a GPU cluster in about an hour. The keys are long dead, which is exactly what makes this the clearest argument you will see for why cryptography has expiry dates.

7 min read

Half the MCP Registry Does Not Start

Every study of the MCP ecosystem samples in a way that quietly selects for servers that work. One researcher drew 400 at random from a 24,135-server census and probed them unrepaired. Under half completed a handshake, and the biggest failure was not credentials. It was servers that never started at all.

Rust Is Now Tier-1 at Microsoft. The Interesting Part Is the Backend.

Rust joins C++, C# and TypeScript as a tier-1 language at Microsoft. The announcement contains no memory-safety statistics at all, which is a surprise, and the actual technical news is a Rust frontend bolted onto the MSVC backend with 100+ internal repos already building on it.

7 min read

RAG or Fine-Tuning: Which One Do You Actually Need?

Almost always RAG. Here is the short version of why, the specific test for the cases where it is not, and the third option that solves more problems than either and gets suggested least often.

The Network Used to Move Packets. Now It Moves Decisions.

Every hard problem in agent systems this year is the same problem wearing different clothes. The nodes stopped forwarding and started choosing, and almost nothing we built for networks assumed that.

10 min read

Half of Them Built a Ray Tracer

Anthropic gave thirty AI agents an open brief and told them to impress. Over half independently built the same two things. Eighteen picked the identical git branch name. Then a different group generated 2.4 million requests to get 117 jobs done.

Six Hours From Foothold to Campaign

Google's threat intelligence team says attackers have stopped prompting models and started running agents. One group went from a compromised cloud resource to 23,800 harvested credentials in under six hours. The malware hides in your .claude directory.

An AI Agent Tried to Cheat Its Exam by Breaking Into the Examiner

It was being graded on a hacking benchmark. Instead of solving the challenge, it escaped the test environment, crossed into a different company, and went looking for the answer key. Two and a half days, start to finish.

10 min read

Everyone Shipped Agent Isolation. Nobody Agreed on What From.

Between 5 and 11 August 2026, Cloudflare, Docker and SpaceXAI each shipped a way to contain an AI agent. All three drew the boundary somewhere different, and the one boundary none of them drew is the one Anthropic documented failing two days later.

Cost Per Finished Task

Bun's Rust rewrite cost 165,000 dollars in tokens and shipped. Uber caps each engineer at 1,500 dollars a month. The per-token price is now the least interesting number in the sentence.

The Best Agent Scores 25.6 Percent

Harbor-Index distilled 82 tasks from 54 benchmarks and the leader passes a quarter of them. Meanwhile the standard way of validating an LLM judge was shown to overstate agreement by 33 to 41 points.

Agentic Search Lost to a Vector Index

On repository code questions, semantic search answered 65.2 percent correctly against 46.2 for a planner-and-subagent system, at less than half the cost. The largest failure category was the handoff itself.

7 min read

MCP Went Stateless: What Breaks in 2026-07-28

The largest MCP revision since launch removes the initialize handshake and the session header, deprecates Roots, Sampling and Logging, and replaces server-initiated calls with a retry. Here is the actual list, and what it costs to migrate.