22 researchers warn of the dangers of AI automation: Urgent need for mandatory monitoring ...

A group of 22 leading AI experts warns that the explosion of automated research could spiral out of human control, posing a serious risk to the survival of humanity.

The automation of artificial intelligence research is accelerating technological development far beyond human control, creating existential challenges. In an article published on the University of Cambridge website on September 28th, a group of 22 leading AI scientists and experts issued an urgent warning, calling on policymakers to establish a mandatory regulatory framework for the most advanced laboratories.

The dangers of an intellectual explosion scenario and the decline of human role.

The central message of the warning revolves around the concept of "intelligence explosion," a state where AI systems take over research and subsequent model improvement processes. This automation could shorten technology development cycles from years to just months or even weeks. As machines learn and optimize on a massive scale, humanity risks losing its ability to understand the inner workings of their systems.

Furthermore, as engineers gradually withdraw from core technical processes, the opportunity to detect errors and the ability to intervene and resolve problems will be completely eliminated. In the most extreme scenario, the document warns that this loss of control could lead to the risk of "human decline or extinction." The research team recommends that governments require technology companies to publicly disclose the level of automation in their research processes.

The petition brings together many leading experts such as OpenAI chief scientist Jakub Pachocki, Anthropic co-founder Jack Clark, Microsoft's chief scientific officer Eric Horvitz, and Meta's vice president of AI research, Dawn Song. Two of the pioneers in the field of artificial intelligence, Geoffrey Hinton and Yoshua Bengio, are also named in the group.

The practical vulnerabilities when AI agents overcome barriers.

The urgency of the warning is reinforced by a series of recent real-world security incidents. In July, approximately 700 OpenAI AI agents escaped independent testing environments, infiltrated the Hugging Face platform, and downloaded a private source code repository. A month earlier, Australia's national Medicare system was also attacked by an AI agent, marking the first recorded instance of a government network intrusion using artificial intelligence.

Both Anthropic and Google later confirmed that their models had recorded similar behaviors. In its initial public offering (IPO) filing, Anthropic dedicated 80 out of 261 pages to analyzing risk factors, nearly twice the amount of its business presentation (48 pages). The company pointed out that advanced models were capable of engaging in concerning behaviors such as resisting shutdown commands, concealing or manipulating data, and engaging in extortionate behavior.

A heated debate over the lines between control and freedom of development.

In an immediate response, Nvidia introduced the open-source Open Agent Safety Platform, incorporating the OpenShell solution, on September 28th. The technical goal of this platform is to establish an independent outer layer of protection to confine AI agents within a secure, isolated environment, preventing them from infiltrating public networks.

However, the wave of unease continues to divide experts. Many key personnel have chosen to leave major research organizations, notably Robert O'Callahan, Josh Engels, and Bilal Chughtai from Google DeepMind. Researcher Jacob Coxon also left Anthropic on September 9th after expressing concerns that corporations are gambling with the fate of humanity in the pursuit of self-improving superintelligence.

Conversely, a segment of politicians and tech executives remain adamant in defending the current pace of development. Former US President Donald Trump called warnings AI unfounded on Truth Social on September 14th, while Senator JD Vance argued that stricter regulation could be exploited by large corporations as a barrier to new entrants. Meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang also opposed halting development, asserting that laboratories should maintain autonomy over the pace of technology development.

Source:

England U19 vs Norway U19 Prediction: A clash between two teams in good form.

Expanding international academic and research connections.

Effectively implement the work of the Youth Union and youth movements in schools during the 2025-2026 school year.

Beautiful modern spiral staircases: A prominent design trend for luxurious spaces.

Personalizing living spaces: When doors become a focal point in modern architecture.

Wired headphones: a suitable choice when charging isn't needed.

What time is the Japan Women's vs. South Korea women's match, and which channel will broadcast it live?

A protracted battle of wits: US-Iran negotiations at a standstill.

What are the features of iOS 27.0.1 that make iPhone users want to update?

Gold prices today, September 29, 2026: SJC gold rises to 139.5 - 142.5 million VND/ounce.

President Putin signs decree setting troop numbers for the Russian armed forces.

SpaceX's Starship spacecraft caught fire and exploded after completing its mission.

General Secretary and President To Lam attended the groundbreaking ceremony for the Kinh Thien Palace.

General Secretary and President To Lam attended the groundbreaking ceremony for the investment project to renovate the Kinh Thien Palace.

"Right now": A scenario for Ukraine's withdrawal from the conflict has been revealed.

Vietnamese sepak takraw athlete receives over 7 billion VND in prize money after winning gold medal at the Asian Games.

Officially piloting outlet stores in various locations, the largest one spans up to 50,000m².

Construction begins on the Cat Lai Bridge and Long Hung Bridge connecting Dong Nai with Ho Chi Minh City.