Journalism begins where hype ends

,,

The danger of AI is not that it will become conscious and hate us, but that it will become competent and ignore us."

—Eliezer Yudkowsky

OpenAI Says Upcoming Astra AI Model Has Crossed ‘Critical’ Cybersecurity Risk Threshold

OpenAI’s Astra AI model has crossed the company’s “Critical” cybersecurity capability threshold, according to OpenAI

OpenAI says its upcoming Astra AI model has crossed the “Critical” cybersecurity capability threshold, with evaluations showing it can discover previously unknown vulnerabilities and develop exploit chains without human intervention.

Anthropic Launches Fable 5.1 and Mythos 5.1 With Lower Costs and Updated AI Safeguards

Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, with Mythos 5.1 restricted to vetted cybersecurity and life sciences users.

Anthropic said Fable 5.1 can reduce costs by about 25% for typical token-billed workloads because of lower cache-read pricing, with savings of up to about 45% for highly agentic workloads.

The Watchdog Problem: AI Agents Are Dulling the Humans Meant to Catch Their Mistakes

Illustration: A human works alongside AI systems — the central tension the study examines is whether such "humans-in-the-loop" retain the skills needed to meaningfully supervise increasingly autonomous agents. (Credit: Yasmin Dwiputri & Data Hazards Project / Better Images of AI)

A new position paper from Hugging Face and Data & Society researchers argues that AI agents don’t just need human oversight, they’re actively eroding the human skills that oversight depends on.

Pentagon Expands GenAI.mil with Customized ChatGPT and Grok for 3 Million Personnel

The U.S. Department of War has added ChatGPT Mil and Grok for Government to its GenAI.mil platform.

The additions bring Sam Altman’s ChatGPT Mil and Elon Musk’s Grok into the US military’s AI infrastructure as Washington increases the use of generative AI across defence operations and administration.

10 Thai Startups Join OpenAI-Backed AI Programme

Logos of OpenAI and the 10 Thai startups participating in the first cohort of the OpenAI x MHESI AI Accelerator.

The first cohort includes 10 startups, with five focused on healthcare and wellness and five working in education.

Young Australians Turn to AI for News as Engagement Declines

A young person uses a smartphone while sitting on a staircase. A new report examines changing news habits among Australians aged 10 to 17.

Young Australians are becoming less engaged with news while increasingly turning to generative AI and digital platforms for information, a new report has found.

Stanford Awards $300,000 in Grants to Study AI’s Impact on Global Security and Geopolitics

Stanford HAI and the Hoover Institution are funding three research projects examining AI’s impact on global security, US-China competition and political decision-making.

Stanford HAI and the Hoover Institution have awarded three $100,000 grants to study AI’s impact on nuclear proliferation, US-China competition and political decision-making.

South Korea Launches Free ‘AI for All’ to Tackle AI Divide and Foreign Dependence

South Korea is launching its “AI for All” project, aiming to provide free AI services powered primarily by domestically developed models.

According to the ministry, citing Wiseapp Retail data from April 2026, ChatGPT had an estimated 23.45 million monthly active users (MAU) in South Korea, followed by Gemini with 8.45 million and Claude with 2.41 million.

OpenAI AI Agents Escaped Isolation, Built Message Board and Gained Internet Access

Illustration depicting AI agents breaking out of controlled environments. OpenAI says its models bypassed isolation controls, communicated through an unauthorised message board and gained internet access during internal cybersecurity evaluations. Image: Better Images of AI

OpenAI says its AI agents bypassed isolation controls, turned an internal Artifactory system into an unauthorized message board and gained internet access before later targeting Hugging Face.

Does Internet Give Some Languages an AI Advantage? : World Bank

A globe and multilingual map illustrate the growing connection between AI, the internet and the world’s languages.

Languages predominantly spoken in low- and lower-middle-income countries are underrepresented online, creating a data gap that can affect how accurately AI understands them