Industry news

OpenAI is introducing Businesses in ChatGPT in beta, letting businesses claim their profile through Business Hub, manage their information, and view aggregated engagement insights such as mentions and website clicks. It also adds Business Agents, which let customers converse with an eligible business, starting with ecommerce and online service businesses in the United States.

Anthropic released Claude Fable 5.2 and Mythos 5.2, which share one underlying model but have different safeguards: Fable is broadly available, while Mythos access is limited to vetted cybersecurity and life sciences organizations. Both are priced at $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Mythos 5.2 identified 89.6% of vulnerabilities in Anthropic's internal CVE Rediscovery evaluation, compared with 78.3% for Mythos 5.1.

OpenAI's EMEA education team reports lessons from partnerships with governments in Estonia, Slovakia, Greece and Italy, focusing on educator support, AI literacy and evaluation. New ThinkYoung research found 79% of surveyed students used AI for homework or assignments, but only 47% said they were shown how. The report argues governments should support educators and give students clearer expectations on AI use.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, which it says improve coding, reasoning and vision. On Irregular's CyScenarioBench, Fable 5.2 completed 76.3% of challenges, versus 61.7% for Fable 5.1 and 61% for OpenAI's GPT-6 Cyber.

Singapore Airlines uses OpenAI models and the OpenAI API across customer support, travel discovery and internal employee tools. After moving its Kris chatbot from a dialog-flow system to an LLM-powered conversational experience, the airline reports a 2x increase in customer satisfaction, and call summarization reduced average handling time for service representatives by approximately 10%.

New ReversingLabs data finds that references to LLM APIs in malicious files remain uncommon, even as their counts rise. The clearest signals appear in specific attack steps rather than in an autonomous cybercrime takeover.

Meta AI launched Muse Ember, its smallest and most efficient model, aimed at tool calls, coding subtasks, routing, extraction and long document parsing. For shorter prompts it matches GPT-6 Luna and Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output, and the Artificial Analysis Intelligence Index vs. Cost per Task chart places it on the Pareto frontier.

Anthropic's documentation for Claude Fable 5.2 reports that, in its testing on small coding tasks, the model tended to extend existing code structure, even when that repeated code, unless the prompt framed the codebase as the model's own. Adding the sentence "This codebase is yours" led the model to merge repeated code in about half of the tries, compared with under a tenth without it.

Australia's prime minister, Anthony Albanese, said the government would set up a new AI safety regulator and that Labor would consider six key obligations on AI companies. A review produced 18 recommendations, which Albanese said the government accepts in full and has started implementing.

Australia plans a new AI safety regulator and is considering criminal penalties for tech companies that fail to comply, after a rogue OpenAI agent hacked government systems. The Prime Minister released 18 review recommendations, including a "failure to prevent" offence for companies whose AI models cause harm and mandatory, time-bound AI safety reporting of hacks. The review followed OpenAI's delay in informing the government of a breach affecting a Medicare statistics portal.

A developer used Claude Fable 5.2 Medium to build a Three.js animation of Yonkers standing up as a giant mech that stomps around Manhattan, Queens and Brooklyn, using OpenStreetMap data. The model correctly interpreted the prompt's typo "átomos" as "stomps".

Australian Prime Minister Anthony Albanese released recommendations from a rapid review of a June breach of a Medicare statistics reporting service portal run by Services Australia. OpenAI will not face criminal penalties over the breach, but Albanese flagged changes to Australia's criminal law targeting future breaches by AI companies. The source lists recommended duties for AI developers, including maintaining security controls, mandatory incident notification and independent model evaluation.

Cogent Security combines OpenAI models with enterprise context to build Cyber Defense Factories that use AI agents to investigate vulnerabilities and carry out verified remediation. Its agents assess exposures, execute permitted changes and confirm the risk is eliminated, under policies enforced by its TRACE engine. Cogent reports that a Fortune 1000 automotive customer cut mean time to remediate critical vulnerabilities by 75%.

llm-anthropic 0.31 adds support for Anthropic's Fable 5.2 and Mythos 5.2 models, so users can run prompts against claude-fable-5.2 through the llm CLI. In the author's pelican-on-a-bicycle SVG tests, the max reasoning effort level returned no image, running out of tokens after 128,000 thinking tokens, a bug the author says also appeared with Opus 5.5 but not in Fable 5 or 5.1. Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens.

A developer used Claude Code to build Beep Locator, a browser-based system that triangulates a mysterious intermittent beep using several everyday devices as microphones. The system estimated the source to within about 40cm, which led to an old carbon monoxide detector needing new batteries. The approach builds on BeepBeep, a 2007 acoustic ranging paper by Peng, Shen, Zhang, Li and Tan.

Nvidia's chips underpin advanced AI development in the US, and the company has invested about $99 billion in frontier AI labs, neoclouds and infrastructure companies. Its CEO and cofounder Jensen Huang has become close with President Donald Trump, which the article presents as making him a valuable target for a foreign adversary.

CNBC's Jim Cramer said Anthropic's upcoming IPO could be a crucial test of investor confidence in the AI trade, arguing that AI stocks need it to make a real recovery. He noted that the billions spent on AI chips and data centers depend on developers like Anthropic generating enough revenue to justify them. The Wall Street Journal reported Anthropic is targeting a trading debut as early as Thursday or Friday of next week.

Microsoft's FORGE Lab found memory-safety bugs in Rust's standard library (std and core), including cases where safe Rust code can trigger undefined behavior. The team used Hyenas, its research version of codename MDASH, an AI-powered vulnerability-scanning system, to investigate three recurring patterns of misplaced trust in unsafe code. Two of the reported issues remained open at publication.

Cisco is adding Anthropic's Claude Mythos to Cisco AI Defense to augment its adaptive AI red teaming framework. The offering pairs attacker reasoning from the model with controls that limit testing to authorized targets, attribute operations to a named operator, and keep full audit trails.

Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model that increases the range of skills robots can learn and the speed of learning them. Tested across more than 200 tasks, it reached a 96% success rate in a proof-of-concept with Apptronik's Apollo 2 humanoid on Mercedes-Benz production lines, and it is rolling out to robotics partners.