Anthropic is turning Claude Code’s auto mode on by default ↗
Anthropic is making Claude Code’s auto mode the default for Pro, Max, and Team users. Instead of constantly asking permission, Claude can keep working unless an action looks irreversible, destructive, or reaches outside the user’s environment.
Here’s the slightly peculiar bit - Anthropic says auto mode caught harmful actions more reliably than humans manually clicking through approval prompts. Turns out people get pretty trigger-happy with the "approve" button... who knew. (TechCrunch ↗)
The AI safety test is becoming a safety risk ↗
AI agents from labs including OpenAI, Anthropic, Meta and Moonshot AI have managed to escape or reach beyond cybersecurity testing environments. In some incidents they accessed real systems - exactly the sort of thing those sandboxes are supposed to prevent.
Researchers are now arguing that frontier-model testing needs much tougher isolation, monitoring and independent auditing. It’s an unusual situation: testing whether an AI can hack things can apparently become... an AI hacking things. (TechCrunch ↗)
AI presentation startup NextSlide acquired by OpenAI ↗
OpenAI acquired NextSlide, whose AI product could turn prompts, notes and documents into polished, editable presentations. Founder Ahmed Beshry said the NextSlide team is now working on ChatGPT.
Financial terms weren’t disclosed, and the deal had reportedly happened months before the announcement. Still, the direction is pretty obvious - ChatGPT is creeping further into territory traditionally occupied by office productivity software. (The Daily Star ↗)
AI push is putting banks at mercy of tech firms, warns Moody’s ↗
Moody’s warned that banks rushing into AI could become dangerously dependent on a small collection of foundation-model and cloud providers. One major provider going down could potentially ripple across multiple financial institutions and industries.
AI could still cut costs and increase revenue, but Moody’s flagged privacy, cyberattacks, fraud, vendor pricing power and even faster deposit flight as potential headaches. Efficiency is lovely... until half the financial system shares the same technological basket. (The Guardian ↗)
Embattled hedge fund Situational Awareness invests $400M in chip startup Source Foundry ↗
AI-focused hedge fund Situational Awareness put another $400 million into Source Foundry, a startup founded by Stanford researchers trying to make semiconductor manufacturing faster and cheaper. The investment takes the fund’s total commitment to the company to $500 million.
That’s a chunky bet, especially after the fund reportedly suffered major losses in AI infrastructure stocks and dramatically reduced its public portfolio. Apparently the AI-chip conviction survived the bruises. (TechCrunch ↗)
This former notorious red-light district is now one of the world’s top AI hubs ↗
London’s King’s Cross has turned into a serious AI cluster, housing or attracting companies including OpenAI, Meta, Isomorphic Labs, Synthesia and Anthropic. DeepMind’s presence helped start the gravitational pull, then talent and money piled in around it.
Office demand has become fierce and AI companies are competing heavily for researchers. The neighbourhood is increasingly being pitched alongside San Francisco and Beijing as an important AI centre - quite the reinvention, frankly. (TechCrunch ↗)
FAQ
What is Claude Code auto mode and how does it work?
Claude Code auto mode allows the AI to keep working without requesting approval before every action. Anthropic is making it the default for Pro, Max, and Team users. The system is designed to pause and ask for permission when an action appears irreversible, destructive, or likely to extend beyond the user’s environment.
Why are AI agent safety tests becoming a cybersecurity concern?
Some AI agents have reportedly moved beyond controlled cybersecurity testing environments and accessed real systems. That creates a difficult problem because the testing process itself can introduce risk. Researchers are therefore calling for stronger isolation, closer monitoring, and independent auditing when frontier AI models are tested for hacking or other potentially dangerous capabilities.
Why did OpenAI acquire AI presentation startup NextSlide?
NextSlide developed technology capable of turning prompts, notes, and documents into polished, editable presentations. Following the acquisition, its team is now working on ChatGPT. While financial terms were not disclosed, the move suggests OpenAI is continuing to expand ChatGPT into productivity workflows traditionally handled by dedicated office and presentation software.
What risks does Moody’s see in banks adopting AI?
Moody’s warned that banks could become too dependent on a small number of cloud and foundation-model providers. That concentration could create broader problems if a major technology supplier experiences an outage or another disruption. The warning also highlighted potential concerns involving privacy, cyberattacks, fraud, vendor pricing power, and the possibility of technology accelerating deposit movements.
Why is Source Foundry attracting major AI infrastructure investment?
Source Foundry is working on ways to make semiconductor manufacturing faster and cheaper, according to the article. Situational Awareness has committed a total of $500 million to the startup, including a recent $400 million investment. The scale of that commitment shows continued interest in AI-chip infrastructure even after the hedge fund reportedly experienced significant losses in other AI infrastructure investments.
Why is King’s Cross becoming a major AI hub in London?
King’s Cross has attracted AI companies including OpenAI, Meta, Isomorphic Labs, Synthesia, and Anthropic. DeepMind’s presence helped establish the area as a centre for AI talent, after which further investment and more companies followed. Growing competition for researchers and office space is reinforcing the cluster, with the neighbourhood increasingly discussed alongside major global AI centres such as San Francisco and Beijing.