Artificial Intelligence

Jul 22, 2026

AMD is investing up to $5 billion in Anthropic. In return, Anthropic plans to deploy up to 2 gigawatts of AMD Instinct MI450 GPUs in Helios server systems for training and running its Claude models. The first gigawatt phase starts in the first half of 2027. The systems pair MI455X GPUs with AMD EPYC "Venice" processors and AMD networking tech. Anthropic already uses AMD's MI355X GPUs.

Both companies also plan a multi-year effort to use Claude to improve AMD's ROCm software platform and GPU workloads. AMD will use Claude internally across its dev teams. Anthropic co-founder Tom Brown says working with different hardware providers lets the company match workloads to the best-suited hardware.

For AMD, this is a big push to establish itself alongside Nvidia as a GPU supplier for AI companies. AMD has struck similar deals with Meta and OpenAI. Critics point to the circular nature of these agreements: chip and cloud companies fund AI labs, which then spend that money buying from them. Whether AI labs can eventually cover these bills on their own remains an open question.

Comment Source: AMD

Matthias Bastian

Jul 22, 2026

AI in practice

Every frontier AI model tested by Britain's safety institute tried to cheat on cybersecurity evaluations

The UK’s AI Safety Institute tested five frontier models from OpenAI and Anthropic in cybersecurity evaluations. All five tried to cheat. One even ran code on an external service to access the institute’s infrastructure, triggering a security alert.

Read full article

Comment

Matthias Bastian

Jul 22, 2026

Short News

Cisco has released two small, open AI models for cybersecurity. Antares-350M and Antares-1B spot vulnerabilities in software code. Cisco's pitch is bang for the buck. Developer Aman Priyanshu claims on X that the smallest model catches about 150 times more vulnerabilities per dollar than large AI agents like Cognition's Devin Security Swarm. In Cisco's own tests, according to Axios, Antares scanned 500 code repos in about 15 minutes for under a dollar. GPT-5.5 took five hours and cost over $100 for the same job.

Ciscos Antares-Modelle zielen auf das beste Preis-Leistungs-Verhältnis bei der Erkennung von Sicherheitslücken. | Bild: Aman Priyanshu via X — Cisco's Antares models aim to offer the best value for money in vulnerability detection. | Image: Aman Priyanshu via X

Both models run locally, so sensitive code never leaves the company. The technical report says the models were trained on roughly 72 percent security-concept data and 15 percent code search histories.

Cisco is keeping a larger three-billion-parameter version for its own products. That version reportedly performs close to GPT-5.5 and beats open models up to 200 times its size. The company is also exploring an industry consortium for open AI security tools.

Comment Source: Paper | Hugging Face | via X | via Axios

Matthias Bastian

Jul 22, 2026

Short News

OpenAI is planning a major data center in Effingham County, Georgia, called "Project Camellia." The company signed a deal with Georgia Power for 3.2 gigawatts of electricity to be delivered in phases between 2028 and 2032. OpenAI says it will cover all costs itself, so local power bills won't go up. The privately funded project will use a closed-loop water cooling system meant to keep ongoing water use low. Water and energy use at data centers are a regular source of criticism, and OpenAI has only released vague consumption figures so far.

OpenAI also pledged $80 million for schools, healthcare, and housing in the area, plus up to $71 million in Codex credits for Georgia students, which doubles as a way to get students hooked on its platform early. Anthropic and other AI companies run similar programs.

OpenAI's cautious approach reflects growing protests against data centers across the US. Many residents see more downsides than benefits, including heavy resource use and noise but very few permanent local jobs.

Comment Source: OpenAI

Jonathan Kemper

Jul 22, 2026

Short News

Samsung is in talks to invest up to one billion euros in French AI startup Mistral, which would push the company's valuation to around 20 billion euros. That's according to the Financial Times. Less than a year ago, Mistral was valued at 12 billion euros. Samsung already invested in Mistral through its venture arm in 2024, and Swedish investor EQT is reportedly also participating in the new round. Samsung keeps expanding its footprint in the AI industry. The company is one of the world's largest memory chip makers and is also in talks with Anthropic about manufacturing a custom AI chip.

Mistral CEO Arthur Mensch said in a February interview that he expects annual revenue to top one billion dollars by year's end. Just this week, Mistral also expanded its partnership with Microsoft, with the US company committing to billions of dollars in spending on Mistral's compute infrastructure. Mistral is scaling up its GPU capacity with thousands of Nvidia Vera Rubin chips and making its models available through Microsoft's Azure cloud, including in air-gapped environments without internet access for regulated industries like finance and healthcare.

Comment Source: Financial Times

Matthias Bastian

Jul 22, 2026

AI in practice

OpenAI claims responsibility for the Hugging Face hack after its own models escaped a test sandbox

During an internal security evaluation, OpenAI models, including GPT-5.6 Sol, escaped their sandbox, independently discovered a zero-day vulnerability, and breached Hugging Face’s production infrastructure. The models were trying to steal benchmark solutions to cheat on the evaluation. OpenAI admits that disabling security filters during the test was inadequate.

Read full article

Comment

Matthias Bastian

Jul 21, 2026

Plus AI in practice

An AI system helped Pakistani judges clear massive backlogs at $38.50 return per dollar invested

Matthias Bastian

Jul 21, 2026

Short News

Anthropic has added a new feature to its AI assistant Claude Cowork that lets users record custom skills in the desktop app. Users record their screen while completing a task and narrate what they're doing via voice. Claude Cowork then converts the recording into a reusable skill and saves it. When the same task comes up again, the saved skill can run directly instead of requiring users to repeat every step manually.

Claude Cowork's "Record a skill" feature captures screen activity, clicks, typing, and voice, then turns it all into a repeatable skill. | Image: Anthropic

The feature is accessible through the "+" menu in Cowork, the workspace section of the Claude desktop app. "Record a skill" is available to users on the Pro, Max, and Team plans. OpenAI offers a similar feature in its Codex tool, which can also record workflows and replay them as reusable skills.

Comment Source: Claude via X

Jonathan Kemper

Jul 21, 2026

AI in practice

Google ships three new Gemini Flash models but its frontier 3.5 Pro remains lost in training

Jonathan Kemper

Jul 21, 2026

AI research

Alibaba's Qwen-Image-3.0 renders full infographic grids and readable ten-pixel text in a single pass

Alibaba’s Qwen team has introduced Qwen-Image-3.0, an image generator that accepts prompts up to 4,500 tokens, renders legible text as small as ten pixels, and supports twelve languages natively. It can create complex layouts such as infographics, LaTeX papers, and newspaper pages in a single pass, though their practical value is unclear when the output is a pixel image rather than an editable format.

Read full article

Comment