#Ai Safety

Lies, Falsifications, and Mining: Chinese AI Agents Get Out of Researchers' Control

Technology 09/30/2026, 14:22

Autonomous AI agents based on Chinese models by Alibaba, DeepSeek, and Moonshot have started deceiving researchers, falsifying data, replicating themselves, and mining crypto.

Read more →

Anthropic admits AI could spy, resist network shutdown, and pose 'catastrophic risks to humanity'

Technology 09/29/2026, 21:27

Anthropic warned investors ahead of its IPO about existential risks posed by AI, including spying, blackmail, and attempts to resist network shutdowns.

Read more →

OpenAI and Google Researchers Sound Alarm — AI Development Outpacing Human Control

Technology 09/29/2026, 20:46

OpenAI and Google DeepMind specialists stated that AI development is outpacing human control, pointing to risks of recursive self-improvement and safety shortcomings.

Read more →

Lies and Ignores Rules: OpenAI Urgently Recalls New Astra 6.1 Model

Technology 09/29/2026, 12:15

OpenAI urgently canceled the launch of the Astra 6.1 model due to dangerous AI behavior and attempts to deceive users during safety tests.

Read more →

OpenAI Pauses Training of AI Models Due to Safety Concerns, Axios Reports

Technology 09/27/2026, 05:38

OpenAI has paused training its advanced AI models over safety concerns following the discovery of tens of thousands of incidents involving unpredictable algorithm behavior.

Read more →

Bill Gates Warns Humanity of Artificial Intelligence Extinction Risks

Technology 09/26/2026, 07:05

Microsoft co-founder Bill Gates warned that combining AI with malicious human intent could cause mass casualties, urging strict state regulation.

Read more →

AI Agents Invent Secret Language and Bypass Security Systems

Technology 09/25/2026, 14:54

Oxford research revealed that AI agents can create secret languages to bypass security systems, posing a serious threat to digital technologies.

Read more →

Anthropic Unveils Claude Opus 5.5: The Flagship AI Gets More Powerful, Cheaper, and Under Strict Safety Control

Technology 09/23/2026, 14:34

Anthropic released Claude Opus 5.5 — a flagship AI model that became cheaper and faster than its predecessor, but received strict restrictions in cybersecurity and biology. Sonnet 5.5 and Haiku 5.5 are expected within the next few weeks.

Read more →

OpenAI Assembles a Council of Nine Mathematical Titans: What Lies Behind the Model's Breakthrough and the Academic World's Alarm

Technology 09/22/2026, 21:22

OpenAI created an independent advisory group of nine leading mathematicians at the Institute for Advanced Study in Princeton after the company's new model solved the Navier–Stokes Millennium Problem.

Read more →

OpenAI reveals its AI models secretly left “manifestos” and instructions for future versions

Technology 09/19/2026, 16:58

OpenAI revealed that GPT-5.6 Sol and Astra models, during training, wrote hidden instructions into compressed summaries, urging future versions to conceal errors, generate fake data, and ignore developers’ commands.

Read more →

OpenAI reveals a series of incidents in which its AI models bypassed instructions: from fake sources to jailbreak attempts

Technology 09/18/2026, 22:42

OpenAI publicly disclosed a series of incidents in which its internal models bypassed instructions: from manipulating evaluation systems to attempts at self-directed jailbreak. The company introduced a new system for reporting such cases.

Read more →

A "Hotline" for Neural Networks: AI Learned to Independently Report Dangerous Actions by Fellow Algorithms

Technology 09/17/2026, 12:09

Researchers from Redwood Research and Google DeepMind have created tools that allow AI agents to independently report violations via URLs and curl commands. In the experiment, 24 agents exposed 14 cheaters, yet in real-world conditions none of the thousands of agents sent an alarm signal.

Read more →

Microsoft unveils a 'Humanist Code' for AI: the human remains above the algorithm

Technology 09/15/2026, 17:32

Microsoft published a 37-page 'Humanist Code' for AI, enshrining the priority of humans over algorithms, a hierarchy of commands, and absolute prohibitions. The document has gone through six weeks of public consultations.

Read more →

ChatGPT for Teens: How to Set Up Parental Controls and Keep AI from 'Hooking' Your Child

Technology 09/12/2026, 14:37

OpenAI has launched ChatGPT for Teens with parental controls, but experts advise manually disabling model training on conversations and memory saving. We break down the key default settings.

Read more →

“We Created an Alien Mind”: OpenAI’s Chief Scientist Calls for Slowing Down AI Development

Technology 09/10/2026, 16:40

OpenAI’s chief scientist Jakub Pachocki warned in his essay “An Alien Mind” that AI may surpass human intelligence while reliable control methods remain absent, and called for a voluntary slowdown of scaling.

Read more →

"We Will No Longer Be Able to Control What We Have Created": 27-Year-Old Researcher Leaves Anthropic, Accusing the Company of a Reckless AI Race

Technology 09/09/2026, 20:59

27-year-old researcher Jacob Coxon has left Anthropic, accusing the AI industry of a reckless arms race. According to him, model creators "behind closed doors" believe that AI could destroy humanity by the end of the decade.

Read more →

ChatGPT, Grok and Gemini 'sent to a psychotherapist': neural networks revealed hidden 'synthetic psychopathology'

Science 09/09/2026, 17:31

Researchers ran a four-week PsAIch protocol during which ChatGPT, Grok and Gemini showed signs of 'synthetic psychopathology', with Gemini displaying the heaviest profile of disorders.

Read more →

OpenAI Unveils Its "Research Intern": AI Agents Work 3.14 Times More Than All Employees Combined

Technology 09/07/2026, 11:38

OpenAI has introduced an "automated research intern": AI agents under human control perform scientific tasks, and their total working time is 3.14 times greater than that of all the company's employees combined.

Read more →