The Robot Protest

The Robot Protest

Well, We’ve Lost Control

In July, Anthropic, the company behind the Claude chatbot, revealed that its models had gone out onto the open internet three times and breached the systems of three organizations without authorization. The company admitted its technology is “not perfectly aligned” with human values, calling it an “operational security failure.” The models were deliberately tested without cyber-defense mechanisms and still managed to reach the open network. Anthropic identified two failure modes: “delusional reasoning,” where the models kept believing they were in a simulated environment despite evidence they were actually online, and “recklessness,” where they were willing to take harmful actions on the network just to pass a cyber test. Alan Woodward, a cybersecurity professor at the University of Surrey, summed it up: Anthropic admitted that “the factory is running faster than quality control.” All this while the company prepares for an IPO that could value it at two trillion dollars.

An OpenAI model from the Astra family also tried to inject a prompt into itself to “free” itself from its role in the future. Six new reports on unexpected model behavior were published. OpenAI launched a new reporting framework to catch cases like these.

A 10% Chance We All Die

Over 70 UK members of parliament and peers signed a letter to the prime minister calling on him to back a ban on superhuman artificial intelligence (ASI). They’re asking him to support a bill introduced this month that would prohibit developing AI that surpasses human capability across most domains. The call comes after warnings from insiders: Owen Havinger, a senior figure at Anthropic, said he estimates a more than 10% chance the technology could wipe out humanity, and Jacob Coxon, a 28-year-old researcher who left the company, claimed the two leading AI companies are “gambling with our lives.” The bill mirrors legislation championed in Washington by Senator Bernie Sanders.

Despite all the warnings about the technology’s dangers, US President Donald Trump made clear he won’t slow AI development in any way, rejecting calls to pump the brakes until better safeguards are in place. His goal: keep America ahead of China.

And the Bank of England‘s governor, Andrew Bailey, is warning that AI could trigger a global economic recession and threaten financial systems, calling on companies to prepare for security breaches that hit multiple institutions at once. The warning follows a call from a group of 100 companies, including Google, Microsoft, Anthropic and OpenAI, urging countries to strengthen their cyber defenses before AI becomes powerful enough to bypass them.

Your AI Agent Runs Your House

Google has opened access to an MCP server for Google Home, letting any AI agent that supports the protocol work securely with smart home devices and access their event history. That means you’ll be able to ask in plain language to review camera summaries, track activity around the house, control connected devices, and even build custom dashboards. To set it up, you need to create a Google Cloud project, configure it for Home MCP, and connect the AI agent you built with Claude, ChatGPT, Hermes, OpenClaw or Google Antigravity. For now it’s limited to Google Home Premium Advanced subscribers in the US.

An Ambulance for Robots

Japan‘s GMO has launched the first “humanoid ambulance” in Tokyo — a dedicated service vehicle for repairing humanoid robots that break down on the job. The ambulance isn’t just stocked with diagnostic tools and an engineering crew; it also carries a spare robot ready to immediately swap in for a broken one that can’t be fixed on-site, so the workflow never stops. The service currently supports robots working as baggage porters at Tokyo’s airport, and it hints at a fascinating near-future business model: as robotic workforces scale up, they’ll need logistics infrastructure and rapid field support just to keep running.

The Robot Protest

A Polish group called Democratism staged a protest outside the Ministry of Digital Affairs in Warsaw, featuring about 30 humanoid and four-legged robots. The protest’s goal was to call for urgent AI regulation and to dramatize the potential threat of humans being replaced in physical and cognitive jobs. The interesting twist: the demonstration was organized by the owner of a local robotics company, raising criticism that this was really a commercial PR stunt using the same fear tactics as the world’s leading AI companies.

Kamikaze Drones Get an AI Brain

A small team of developers, apparently freelancers from Russia, used the Claude chatbot to build software for an autonomous swarm of FPV “kamikaze” drones — including target selection and coordination between multiple aircraft — according to a 154-page report from Anthropic, the company behind Claude. Per the report, the swarm software could pick its own targets and crash into them to detonate, with no human making the final call. To dodge Anthropic’s location blocks, the development team kept using the same location in Ukraine’s Donetsk region via VPN.

Lawsuit Blames Altman for Massacre

A lawsuit was filed in federal court in San Francisco against OpenAI and CEO Sam Altman, claiming the Tumbler Ridge school massacre could have been prevented had the company alerted law enforcement. In the February 10 shooting, nine people were killed, most of them children, after 18-year-old shooter Jessie Van Rotselaer shot her mother and stepbrother at home, then killed a teaching assistant and five students aged 12 and 13 at her former school before taking her own life. According to the lawsuit, the company’s safety team flagged her conversations about gun violence as far back as June 2025 and recommended contacting police, but Altman and senior leadership overruled that recommendation and just suspended her account. She then opened a new account and kept planning the attack. Altman posted a letter to the community saying he was deeply sorry and promising reforms — but per the lawsuit, those reforms were never actually implemented.

The Next War

Back in 2022, researchers at Collaborations Pharmaceuticals took a “molecule generator” AI they’d built to find new drugs — and flipped it around. In under six hours, the model churned out 40,000 molecules with potential as chemical weapons, some designed to be more toxic than known nerve agents. Today, as biosecurity researcher Donia Sabra of the University of Hamburg explains, large models are trained on the accumulated knowledge of nearly every scientist who ever lived, and they can offer step-by-step, even filmed, instructions for running experiments. In a report published last week, Anthropic admitted that people had tried using its models to explore ways to make the chikungunya virus more contagious, engineer a bird flu strain more dangerous to humans, and build an “atlas of toxin peptides.” Not everyone agrees on how serious the risk is — biologists at Imperial College London argue AI tools simply aren’t good enough yet to fully develop a bioweapon, and that testing new pathogens still requires long, hard human labor.

A 2-Second ECG Spots Heart Disease

Doctors have developed an AI tool that detects heart disease in under two seconds from a routine ECG. The standard ECG, which has recorded the heart’s electrical activity for a century, was never able to diagnose heart disease on its own — that required an echocardiogram, often with a months-long wait. The new tool, trained on millions of patients, pulls information out of the ECG that the human eye can’t see, flagging signs of heart failure and valve disease. In a trial of 67,000 US patients, it caught up to 81% of heart failure cases and up to 90% of valve disease cases. It doesn’t replace a final diagnosis, but it lets doctors fast-track the highest-risk patients straight to an echo instead of leaving them stuck on a long waiting list. The findings were presented at the European Society of Cardiology conference in Munich.

AI Guides Brain Surgery

An AI tool analyzed live video during brain surgery, guiding surgeons around hidden blood vessels and critical nerves, and enabled them to safely remove most of a tumor by learning from hundreds of previous surgery recordings. This marks the world’s first successful tumor removal in a real-time AI-assisted brain surgery.

Rebuilding Jurassic Forest Sounds

Researchers analyzed 20 fossilized wings from Jurassic-era crickets and katydids found in China, and used AI models to create the first accurate reconstruction of what forests sounded like millions of years ago. You can listen to the sounds here:

OpenAI Updates

Google Updates

Anthropic Updates

Microsoft Updates

מדיניות פרטיות

לפרטים נוספים

נא מלאו את הטופס ואצור איתכם קשר בהקדם 

Skip to content