OpenAI and Oracle to Build ChatGPT into Oracle Business Systems: Today's Other News

OpenAI and Oracle are expanding their partnership. Oracle is building the Oracle AI Workspace platform, which connects corporate databases and Oracle business logic to the ChatGPT interface. Employees no longer need to open reports and export spreadsheets: they type a request in plain language and get an analysis of financial figures or launch a workflow. This is the main story of the day. Four more came out alongside it, and each is covered below.

Oracle connects ChatGPT to corporate data

The data stays inside Oracle's protected environment. Corporate policies and access rights remain in force: ChatGPT shows a user only what that person is already allowed to see.

The relationship runs both ways. Oracle itself uses OpenAI models in two new products:

• AI Security Mesh. Automatically searches systems for vulnerabilities. • App Composer. An environment where Codex, OpenAI's coding agent, builds a company's internal services.

There is a commercial side too. Oracle customers will be able to buy access to the OpenAI API and enterprise versions of ChatGPT through the Oracle marketplace and pay with cloud credits. A company that already has an Oracle Cloud budget needs no separate contract or separate invoice. For OpenAI, this is a sales channel into large enterprises. For Oracle, it is a way to keep customers who want AI next to their data.

Lawyers, a model ranking and cybersecurity: three releases

Anthropic: artifacts in Claude for Legal. Claude for Legal now has a gallery of ready-made interactive tools for lawyers. The artifacts check citations, compare contracts against corporate standards, build summary tables across jurisdictions and track changes in regulations. Users do not have to upload files manually: Anthropic added more than 60 connectors to familiar legal databases, and the model attaches a link to the primary source in its answer. PwC, Goodwin, Harvey and others have already built their own artifacts. Separate tools were made for legal aid organizations: they process immigration documents and look for errors in rental registries.

Artificial Analysis: Intelligence Index v5. Artificial Analysis tests models independently and combines the results into a single ranking. In the fifth version, 60% of the index is built on closed task sets, so models cannot memorize the answers. Two tests were added:

• FrontierCode 2.0. Programming tasks. • Terminal-Bench-Science. An environment where autonomous agents solve scientific problems.

The scenarios got harder: the median number of generated tokens (the chunks of text a model assembles its answer from) rose by 66%, because agents spend a long time computing. Gemini 4 Carbon leads the ranking, followed by Claude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol. The gap between closed and open models has widened. The best open model, DeepSeek V4.1 Flash 1017, scored 32 points, 27 fewer than the leader. That puts the leader at about 59.

Microsoft: MAI Cyber 1.1 Flash. The model is built into the MDASH security platform and analyzes source code. In the multi-agent testing system, it handles the bulk of bug searching and verification and passes only non-trivial cases to larger models. Scanning got nearly a fifth cheaper, and detection quality did not drop. According to Microsoft's statistics, customers who run an audit through MAI Cyber once stay on this profile 90% of the time, rather than on a combination of OpenAI models. The company's own AI accounts for about half of the checks in MDASH. These figures come from Microsoft itself, and there is no independent verification yet. Access is open to vetted users of the platform.

SpAItial Echo-3: a room scanner in your smartphone

SpAItial released Echo-3, a world model that turns phone footage into a detailed three-dimensional space at real-world scale. The SpAItial iOS app launched alongside it. It digitizes a space, from an apartment to a factory, in a couple of minutes, with no tripods or extra equipment.

The model's main feature is that it fills in blind spots. The network completes areas missed during filming based on context, keeping metric accuracy and geometry free of the artifacts common in other scanners. The app is already in the App Store, the first month is free, and an Android version is promised later. Echo-3 will be opened to enterprise customers through an API. Precise dimensions will be useful in real estate, simulations and spatial analysis by AI agents.

What this means

All five stories point the same way: AI is moving into ready-made workflows. Oracle connects ChatGPT to databases, Anthropic builds tools for lawyers, and Microsoft hands routine bug hunting to a cheap, narrow model. For engineers and IT directors, the practical takeaway is to look not at the ranking but at where a model gets access to data and who is responsible for permissions. The same ranking also carries a warning: open models currently lag far behind the leaders on complex agentic tasks.