OpenAI’s system combines broad activation probes, a model-based reviewer, investigative agents, system telemetry and human response. The company says it monitors frontier training runs, evaluations and internal Codex traffic. Some external deployments can pause a conversation or send an alert, but coverage varies by product and conversation-tracking method.
Artificial intelligenceOpenAI’s system combines broad activation probes, a model-based reviewer, investigative agents, system telemetry and human response. The company says it monitors frontier training runs, evaluations and internal Codex traffic. Some external deployments can pause a conversation or send an alert, but coverage varies by product and conversation-tracking method.
Read the article
Artificial intelligenceModern biology produces more data than human teams can inspect line by line. On October 21, OpenAI described work on a whole-brain CRISPR atlas. Each GPT-Rosalind-powered agent studied one gene disruption, examined cellular responses, searched the literature and proposed an explanation. The findings were combined into a map linking genes, cells and publications.
Read the article
Artificial intelligenceA general agent can write, search and use tools, but may not know a company’s procedure. A skill fills that gap with a folder containing a `SKILL.md` file, instructions and optional scripts or references. Anthropic says it created the format for Claude Code, opened it in December 2025 and transferred its name, logo, domain and GitHub organization to the AAIF on October 22, 2026. Its claims of more than 70 integrations, one million published skills and nearly 280 million installations need independent measurement.
Read the article
Artificial intelligenceA model can quickly make a chart. Linking a selection to another chart, table or paragraph is much harder. Released on October 22, Flint-interactive adds semantic events and update operators to Flint. The agent states the intended interaction and the compiler handles part of the technical logic.
Read the articleArtificial intelligenceOpenAI’s system combines broad activation probes, a model-based reviewer, investigative agents, system telemetry and human response. The company says it monitors frontier training runs, evaluations and internal Codex traffic. Some external deployments can pause a conversation or send an alert, but coverage varies by product and conversation-tracking method.
Read the articleArtificial intelligenceModern biology produces more data than human teams can inspect line by line. On October 21, OpenAI described work on a whole-brain CRISPR atlas. Each GPT-Rosalind-powered agent studied one gene disruption, examined cellular responses, searched the literature and proposed an explanation. The findings were combined into a map linking genes, cells and publications.
Read the articleArtificial intelligenceA general agent can write, search and use tools, but may not know a company’s procedure. A skill fills that gap with a folder containing a `SKILL.md` file, instructions and optional scripts or references. Anthropic says it created the format for Claude Code, opened it in December 2025 and transferred its name, logo, domain and GitHub organization to the AAIF on October 22, 2026. Its claims of more than 70 integrations, one million published skills and nearly 280 million installations need independent measurement.
Read the articleArtificial intelligenceA model can quickly make a chart. Linking a selection to another chart, table or paragraph is much harder. Released on October 22, Flint-interactive adds semantic events and update operators to Flint. The agent states the intended interaction and the compiler handles part of the technical logic.
Read the article
Artificial intelligenceOn October 21, the companies announced the first private-preview access for selected enterprise customers. Astra targets demanding tasks, Sol balances capability and cost, and Luna targets fast work at scale. This does not merge ChatGPT and Gemini; it gives IT teams another route to call OpenAI models, compare them with other offerings and connect Codex tools to Google’s Gemini platform for inference.
Read the article
Artificial intelligenceMarkets await Alphabet and Amazon results to see whether AI revenue is beginning to match data-centre spending.
Read the article
Artificial intelligenceMAI-DxO organises questions and tests as a clinical pathway. Its results are strong, but hospital validation is only beginning.
Read the article
Artificial intelligenceAI can sort many images, but it cannot turn a silhouette in a crowd into judicial truth. Its strongest role concerns collective, anonymized situations.
Read the article
Artificial intelligenceOn 13 October, Google announced Gemini 4 Argon’s availability for developers and businesses, alongside a rollout in its paid offerings.
Read the article
Artificial intelligenceLaunched on 7 October, Claude Haiku 5.5 targets high-volume, cost-sensitive tasks, according to Anthropic.
Read the article
Artificial intelligenceOn 16 October, Beijing opened a consultation on export controls covering, among other areas, 23 AI-related technologies.
Read the article
Artificial intelligenceOn 15 October, Anthropic announced broader internal access for a team from independent evaluator METR.
Read the article
Artificial intelligenceReuters reported a sharp fall in CoreWeave bonds on 16 October.
Read the article
Artificial intelligenceCERN and Mistral want to adapt AI to particle collisions. The scientific ambition still needs demonstrable results.
Read the article
Artificial intelligenceMistral plans to open Large 4 on 27 October. Companies will still need to assess cost, security and licensing.
Read the article
Artificial intelligenceAI investment supports growth, but countries, occupations and regions do not have equal capacity to benefit.
Read the article
Artificial intelligenceAI is absorbing some junior tasks. The task is not only to teach the tool, but to preserve the first step of a career.
Read the article
Artificial intelligencePublic pre-release tests can uncover risks earlier, but the reported US-UK restart still needs official confirmation.
Read the article
Artificial intelligenceS&P 500 earnings may jump 31%, while technology and three AI giants concentrate almost two-thirds of the gain.
Read the article
Artificial intelligenceThe new Gemini agent can act inside workplace tools. Its value will depend on control over access, errors and decisions.
Read the article
Artificial intelligenceAnnounced on 1 October, MAI-Transcribe-2-Streaming turns incoming speech into text as it listens. This development could make meetings and classes easier to follow, but Microsoft says the service remains a preview, without a production availability guarantee.
Read the article
Artificial intelligenceCloudflare launched Clef and Clef-flash on 1 October. These models select among permitted answers and assign probabilities to them. For businesses and customers, the challenge is practical: routing a request to the right service without mistaking a score for a fair decision.
Read the article
Artificial intelligenceBlack Forest Labs added FLUX 3 Image to its offering on 1 October. The model can target elements within an image and combine several references. It could save creative time, with an explicit limitation: shadows, reflections or nearby lighting can still change.
Read the article
Artificial intelligenceResearch published on 24 September explores a hiring paradox: making applications easier can make skills harder to identify. Its theoretical model suggests a growing advantage for already visible experience, without demonstrating that this scenario occurs in every business.
Read the article
Artificial intelligenceOn 1 October, Microsoft confirmed closer integration between Copilot and Office. The aim is to make documents part of an assisted workflow while retaining control.
Read the article
Artificial intelligenceAMD plans to acquire World Labs to combine processors, software and spatial intelligence for robotics and 3D environments.
Read the article
Artificial intelligenceSince 2 August 2026, EU obligations require users to be informed and some artificial content to be identifiable.
Read the article
Artificial intelligenceWith Opus 5.5, Anthropic is betting on faster, less expensive agents for coding, tool use and complex professional work.
Read the article
Artificial intelligenceAutonomous agents, more efficient models and new uses: the advances worth following without giving in to the hype.
Read the article
Artificial intelligenceWhat AI agents can do, their limits and the essential precautions.
Read the article