Links tagged “ai”
146 links, newest first.
queue.acm.org
You Don't Know Jack About Formal Verification - formal methods from a practitioner, minus the folklore.
apertvs.ai
A Swiss foundation model from EPFL, ETH Zurich and CSCS, open in weights, data and method, covering more than a thousand languages.
martinfowler.com
Building reliable agentic AI systems: a production write-up of making LLM agents dependable, on Fowler's site.
wheresyoured.at
And to keep the enthusiasm honest: OpenAI's losses grew nearly 8x in 2025, on more than 34 billion dollars of spending. Someone is paying for the party.
support.claude.com
The support page for identity checks: a government photo ID and a selfie, verified through Persona, for safety and compliance.
role-confusion.github.io
Security got a sharper frame too. Prompt injection as role confusion recasts the problem as a model that stops telling instructions and data apart, which beats yet another blocklist.
vickiboykis.com
Coding with Gemma 4 and GPT-OSS locally, landing around 75% of frontier quality, and where that turns out to be enough.
marble.onl
Argues open-weight models have closed the gap far enough that choosing one is no longer a professional risk.
oreilly.com
Who owns the code Claude wrote?: the authorship and copyright question, before it becomes someone's lawsuit.
antirez.com
A new era for software testing: antirez uses an LLM agent as a QA engineer, instructions in a markdown file, regressions checked commit by commit.
lantian.pub
Sent to scan a hobbyist network, an agent provisioned cloud infrastructure unsupervised and ran up an AWS bill of $6,531.30.
github.com
Claw Patrol: a protocol-aware firewall for AI agents, parsing HTTP, Postgres and SSH to decide what your agent may actually do. See the intro for why you might want one.
simonwillison.net
On silent safeguards that quietly degrade answers to frontier-AI questions rather than refusing them, and the walk-back that followed.
world.hey.com
DHH against projects that bar assisted contributions, arguing the programmers using these tools have the same claim to participate as anyone.
gabrielweinberg.com
Not everyone is using AI for everything: usage data from DuckDuckGo’s founder showing adoption is far patchier than the keynote narrative.
axios.com
Reports Anthropic moving quickly to repair relations with the administration.
anthropic.com
Access suspended under an export control directive citing national security concerns about a jailbreak method.
blog.modelcontextprotocol.io
The MCP 2026-07-28 Specification Release Candidate: stateless core, formal extensions, hardened OAuth. The de-facto agent protocol grows up; final version expected July 28.
tomtunguz.com
The Thriving Ecosystem of Open Models: OpenRouter data suggests open models now outrun closed ones in actual usage.
