Industry news

OpenAI and Oracle are expanding their partnership to bring Oracle business applications and data into ChatGPT, with access governed by each organization's permissions and approvals. Oracle is developing Oracle AI Workspace, Oracle AI Security Mesh, Oracle AI SMB Bundle and Oracle App Composer, all built with OpenAI models and tools. U.S. Oracle customers can buy access to the OpenAI API, Codex and ChatGPT Work through Oracle Marketplace.

Anthropic has filed to list on the US stock market, and its S-1 prospectus is the first public look at audited financials. The filing, released Monday, lists more than 90 pages of risks, from cyber-attacks to the limits of its safeguards. The source describes the business as explosive but very expensive growth.

A growing set of benchmarks measures AI model security, but each targets different variables. None of them replaces testing against your own code and workflows.

Check Point uses OpenAI Daybreak to cut its vulnerability research workflow from hours to minutes. Its analysts turn vulnerability findings into security signatures or indicators of compromise that protect over 100,000 customer organizations. Check Point is also building an Agentic Network Security Orchestration solution with Daybreak, with human approval for consequential actions.

OpenAI released GPT-6.1 Terra, a fast, efficient model it positions for visual reasoning, computer use and clearer writing. The model is priced at $1 per million input tokens and $5 per million output tokens, about half the price of GPT-5.6 Terra and GPT-6.1 Sol and around 10x the price of GPT-6 Luna.

OpenAI has announced three new scientific results in mathematics, computer science and physics, less than a month after releasing hundreds of manuscripts it said contained solutions to long-standing math questions. For the computer science and physics results, external experts it calls "shepherds" guided the work, while for the math problem they advised OpenAI to release the AI output so researchers could work through their own explanations.

Anthropic filed for an IPO that could be the largest initial stock offering in history, with investors measuring its potential market cap in the trillions. The company's business is deeply tied to tech megacaps, with close to half its revenue coming from Claude sold through Microsoft, Amazon and Alphabet cloud divisions.

Endor Labs has published The Applied AI Playbook for AppSec. The playbook describes four levels for building security into the agentic software development lifecycle, covering the code that coding agents write and the use of AI within an organization's own AppSec program.

Endor Labs has published a solution brief on Package Firewall, which sits between developers, AI coding agents and public registries. It blocks malicious, vulnerable and policy-violating packages at install time, before they reach a workstation or CI pipeline. The brief notes that every install pulls code from the internet directly into the environment, and that no post-install scan can undo that.

Endor Labs explains AI code review, which uses large language models and static analysis to read pull request diffs, flag security, logic and design risks, and post findings as PR comments before merge. The article cites a Georgetown CSET test where almost half of the code snippets from five large language models contained security weaknesses, and it argues that human reviewers remain the final decision makers.

Greptile, which builds AI agents that review and test code changes, describes how it uses GPT-6 Astra's computer use to automate application testing at scale. Its agents start a development server for a pull request, resize the browser to an iPhone-sized viewport, and simulate user workflows to catch problems such as a button that disappears. The company also says it indexes customer codebases and serves that documentation to Codex through an MCP server, and tracks AI-generated code reverts as a measure of impact.

Anthropic's S-1 filing with the US Securities and Exchange Commission acknowledges that its data centers increasingly rely on natural gas, which is primarily methane. The company says this leaves its AI data center buildout exposed to rising fuel costs and existing supply chain constraints.

Researchers recovered eight eMudhra intermediate CA private keys, which the source says were supposed to be 384-bit but fit in eight bits, so every possibility can be tried by counting to 255. eMudhra attributes the flaw to a key-generation ceremony in April 2025 and says the intermediates never issued customer certificates. Browsers blocked the certificates, while other clients needed a new bundle or verifier change.

OpenAI has launched GPT-6.1 Terra, a model focused on computer use, visual reasoning, coding, and everyday work, priced at $1 per million input tokens and $5 per million output tokens. OpenAI says it is 50% cheaper than GPT-5.6 Terra and GPT-6.1 Sol, and scored 15 percentage points higher than GPT-5.6 Terra on Agents' Last Exam at medium reasoning effort. It is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-terra.

Anthropic filed papers for an initial public offering that could be the largest ever, planning a Nasdaq listing under the ticker "ANT" next month. The 335-page filing reports revenue above $16 billion in the first half of the year, 14 times the same period last year, and a net loss. The source also notes that the CEO has warned advanced AI could pose an existential threat to humanity.

Cisco is launching Model Foundry, a platform service in Cisco Cloud Control that gives partners access to Cisco's purpose-built models and lets them serve those models in customer applications. Partners can also fine-tune the models on their customers' own data, with built-in evaluation comparing base and tuned models side by side. The platform starts with models including Deep Network 1.4, Antares-3B, Decision Model 1.0 and Cisco Time Series Model 1.0.