Yesterday we told you to patch PHP. Today the bigger story is how those bugs became dangerous: AI models turned flaws hidden for a decade into a working exploit, and governments, hospitals and companies switched their websites off. Also: a new AGI forecast from Demis Hassabis, Tesla’s first production Optimus robots, a record round for DeepSeek and robots that now paint and sew.
Watch the 42-second short:
Also on Telegram: t.me/arniron/35
1. AI found the PHP exploit path, and sites went dark
Security firm Wiz says the two critical PHP flaws disclosed on 21 October sat in the codebase for more than a decade and would have been extraordinarily hard for a human to exploit. Its research used its own autonomous tool Wiz Atlas plus Google’s Gemini 4 Argon and OpenAI’s GPT-5.6 Sol. “Pre-AI, this manual work would be difficult to implement and even harder to scale,” Wiz wrote. “With AI this is no longer a barrier.”
The reaction was unusual. Germany’s ITZBund, France’s prefecture websites, the Port of Rotterdam, Dutch hospitals, the Czech government, Finland’s Valtori, Wikimedia and BBC portals went offline as a precaution, and in Japan several ministries plus Nomura, MUFG, ANA and JR East suspended some services; Prime Minister Sanae Takaichi urged site owners on X to apply the fix. Most operators reported no evidence of a breach. The caveat: the patch is public now, and experts expect attackers to use AI to reverse-engineer it. If you run PHP, update today.
2. Hassabis: AGI in 2–3 years. Hinton: AI shows signs of consciousness
Google DeepMind CEO Demis Hassabis told The Wall Street Journal that artificial general intelligence could arrive within roughly two to three years. At the start of 2026 he still spoke of five to eight years. “I keep shifting my predictions closer and closer,” he said.
Geoffrey Hinton went further at the Starmus festival in Tenerife, as reported by The Telegraph: today’s AI already shows signs of consciousness, and humanity has to make it genuinely care about people. “If we don’t, we’re finished.” Both are forecasts and opinions, not measurements; there is no agreed test for AGI or machine consciousness. But when two Nobel laureates move their clocks, the debate on safety moves too.
3. Robots take up the brush and the needle
A Kinova robot arm, originally designed as an assistive arm for people with disabilities, was given a neural network and now paints with brush and acrylics in a style that looks professional. Our short shows it at work in a gallery, layering thousands of small strokes into a mountain landscape.
In China, Aitu, part of industrial sewing-machine maker Jack Technology, tested sewing humanoids at garment factories including down-jacket maker Bosideng and is building up production for up to 10,000 robots a year. Aitu itself says humanoids are “still at the learning stage”: they load fabric and feed machines, while fixtures do the stitching. Still, the clouds over the seamstress profession are gathering.
4. Tesla starts series production of Optimus
The first production-standard Optimus humanoids have come off the line at Tesla’s Fremont plant in California, which used to build Model S and X cars. Tesla wants 20,000 robots a week by the end of 2027. “We’ve proven the current Optimus production technology fits a commercial product,” Elon Musk said.
The caveat is the ramp: reliability is still the main obstacle, Musk admitted, and a supply chain for 20,000 units a week does not exist yet. Tesla’s automotive profit fell in the third quarter, so a lot is riding on the robot.
5. DeepSeek raises about $15bn
Chinese AI lab DeepSeek has raised around 100 billion yuan (about $15bn), Bloomberg reported, citing people familiar with the deal. Tencent and battery giant CATL are among the investors. The company originally targeted about 50 billion yuan and doubled the round on demand, making it one of the largest private financings in China.
The money is meant for compute, including a large data centre in Inner Mongolia, to compete with OpenAI and Anthropic. A listing on Shanghai’s STAR Market is being prepared for 2027. DeepSeek has not confirmed the figures.
6. Berlin’s Cygrid raises €2.66M to keep AI agents in check
Berlin start-up Cygrid has come out of stealth with €2.66 million. Its pitch: today’s identity security checks who logs in, but AI agents keep acting long after the login. Cygrid checks what an agent may do throughout the session, action by action, so it cannot wander into confidential company data.
The company wants to be a European alternative to US and Israeli identity-security vendors and will use the money for enforcement features, hiring and sales. It is an early-stage bet; the category of agent security is crowded and still defining its standards, from MCP to the Linux Foundation’s Agentic AI Foundation.
7. Microsoft’s diagnosis AI beats GPs on hard cases
A peer-reviewed paper in Science describes Microsoft AI’s diagnostic orchestrator (MAI-DxO), which works through a case step by step, ordering tests and asking questions like a panel of doctors. On 300+ difficult cases from the New England Journal of Medicine it reached the right diagnosis more than twice as often as general practitioners.
Caveats matter: these were historical, published cases, the doctors were generalists working alone, and the costs were modelled. Microsoft plans to test the tool at a Boston hospital next. It is a research result, not a product in clinics.
Sources
- RTÉ News — Websites go dark after AI-discovered flaw sparks security scramble
- The Korea Times (Reuters) — Govts, companies take websites offline over PHP vulnerabilities
- ABC7 — Websites shut down globally after AI finds critical PHP flaw
- php.net — security releases
- Hightech+ — Демис Хассабис: до AGI осталось 2-3 года
- Hightech+ — Хинтон: ИИ уже обладает признаками сознания
- Hightech+ — Маск начинает серийное производство Optimus
- Investing.com (Bloomberg) — DeepSeek raises $15 bln in Tencent, CATL-backed round
- EU-Startups — Berlin's Cygrid exits stealth with €2.66 million
- Microsoft AI — The path to medical superintelligence (MAI-DxO)
Leave a Reply