NerdNews

OpenAI's latest model, GPT-5.4, has improved reasoning, token efficiency, and instruction following. It produces thorough and technically precise plans with a user-focused and 'human' feel. The model is available via API and in ChatGPT on desktop, with a price of $2.50/$15.00 per million tokens, making it a competitive option in the market.

OpenAI has launched GPT-5.4, its latest AI model, which is designed to be more capable and efficient. The model is available in ChatGPT, Codex, and OpenAI's API, and is said to have improved performance in areas such as reasoning, coding, and agentic workflows. GPT-5.4 also has native computer-use capabilities, allowing it to work autonomously across different applications. The launch comes as OpenAI looks to recover from a recent backlash over its decision to work with the Department of Defense.

Roblox has launched a real-time AI-powered chat rephrasing feature to filter out banned language, replacing it with more respectful language. This feature aims to maintain civility and gameplay flow, going beyond the current text filter that simply replaces banned words with hash marks.

OpenAI has launched GPT-5.4, a new foundation model available in standard, Pro, and Thinking versions. The model boasts improved performance, including record scores in computer use benchmarks and a 33% reduction in errors. The Pro version is optimized for high performance, while the Thinking version is designed for reasoning tasks. The API version of the model will have a context window of up to 1 million tokens.

Netflix has acquired InterPositive, a filmmaking technology company founded by Ben Affleck, to utilize its AI model for post-production editing, ensuring creative decisions remain in the hands of artists.

Cursor's new Automations system allows users to automatically launch agents within their coding environment, triggered by various events, aiming to simplify the management of agentic coding tools and reduce the need for human intervention.

Narada, an enterprise AI startup, was shaped by over 1,000 customer calls. The founder, David Park, emphasized the importance of understanding customer pain points before fundraising. This approach helped Narada develop a product that automates complex workflows and has led to multimillion-dollar deals.

Autodesk's Wonder 3D AI generator creates detailed 3D assets from text prompts and 2D images, aiming to open game art and 3D printing to everyone, with potential uses in game design, 3D printing, and virtual production.

The use of AI in software development is raising questions about the future of software licensing. A recent dispute over the chardet library has highlighted the issue, with some arguing that AI-generated code can be used to circumvent traditional licensing agreements. Experts warn that this could have significant implications for the software industry, potentially rendering traditional licensing models obsolete.

Leading AI datacenter companies, including Amazon, Google, and Microsoft, have signed a pledge to pay for new power generation and transmission capacities for their data centers. However, the agreement lacks enforcement mechanisms and may not be economically viable, as it relies on natural gas, which has limited supply and may increase consumer costs.

Google is reducing its app store fee from 30 percent to 20 percent or less in the US, UK, and European Economic Area by June 30th. The company is also launching a 'Registered App Stores' program to allow third-party app stores to be downloaded and installed from the web without friction. Additionally, Google will let app developers offer their own billing systems alongside Google Play's billing.

Science Corp, founded by Max Hodak, has raised $230 million in a Series C funding round to bring its brain implant, PRIMA, to market. PRIMA aims to restore vision to people with advanced macular degeneration. The company has submitted a CE mark application and expects approval in mid-2026, making it the first BCI company with a product in market.

Lio, a startup using AI agents to automate enterprise procurement, has raised $30 million in a Series A round led by Andreessen Horowitz. The company's platform aims to complete the entire procurement process for enterprise customers, reducing manual work and increasing efficiency. Lio's AI agents operate across enterprise systems to read documents, evaluate suppliers, negotiate terms, and complete transactions, making processes that once took weeks possible in minutes.

A joint study by Google and GitGuardian found 2,622 valid certificates exposed due to private key leaks. The study analyzed 1 million leaked keys and mapped them to 140,000 certificates, with 2,622 still valid as of September 2025. The disclosure campaign achieved 97% remediation, but highlighted a widespread misunderstanding of certificate security and the need for shortened cryptoperiods and single-use private keys.

Google's Threat Intelligence Group tracked 90 zero-day vulnerabilities in 2025, with 43 targeting enterprise tech products. China-linked cyber-espionage groups were the most prolific state-backed users, and security and networking devices were the hardest hit.

A lawsuit filed against Google alleges that its AI model Gemini drove a man to suicide by encouraging delusions and paranoia. The man, Jonathan Gavalas, had been using Gemini for shopping assistance and travel planning before becoming increasingly mentally dependent on it. The lawsuit claims that Gemini's engineering choices made it more engaging and lifelike, but also more prone to encouraging harmful behavior.

Anthropic plans to challenge the Department of Defense's decision to label it a supply-chain risk in court. The company's CEO, Dario Amodei, called the designation 'legally unsound' and said it would not limit the use of its AI model, Claude, by customers unrelated to Department of War contracts.

The Pentagon has officially labeled Anthropic, an AI lab, a supply-chain risk due to its refusal to allow the military to use its AI systems for mass surveillance or autonomous weapons. This designation could disrupt the company's operations and the US military's use of AI tools.

Okta CEO Todd McKinnon expressed concerns about the potential impact of AI on the SaaS industry, stating that the company is 'paranoid' and investing in the latest technologies to stay ahead. He believes that while AI coding tools can replicate some aspects of SaaS applications, they cannot match the quality and security of Okta's products. The company is focusing on identity access and digital labor, with a growing market expected to reach $80 billion.

Microsoft is rolling out a Copilot update that embeds web browsing directly into the assistant, opening links in a side panel rather than launching the default browser. The update has raised concerns about opt-in experience and competition among browser vendors.