#Safety & alignment

The surge in data centers in Texas could wipe out decades of air quality improvements.

The surge in data centers in Texas could wipe out decades of air quality improvements. HARC reports show that if all planned projects are connected to the grid and rely on the expected mixed energy mix, fine particulate matter pollution will exceed 2006 levels within five years, NOx emissions will increase by 136%, and CO2 will increase by 90%. The existing and under-construction data center capacity in Texas reaches 26GW, and the planned self-consumption capacity is the highest in the United States.

Patch the Planet: Global Security Vulnerability Response Coordination Platform Launch

Patch the Planet connects security researchers, maintainers, and incident responders to help address critical vulnerabilities in open-source projects.

Wiz Research uses AI to discover PHP core vulnerability PHP2Shell

Wiz Research has discovered multiple severe memory safety vulnerabilities in PHP using cutting-edge AI models, allowing unauthenticated attackers to remotely execute code or leak memory-sensitive data, affecting PHP versions from 5.4 to the latest; the vulnerability chain requires complex steps, and AI has significantly reduced the difficulty of its implementation, being considered one of the most severe vulnerabilities in software history, PHP team has released a fix patch.

NVIDIA Confidential Computing expands, providing secure extension capabilities for cutting-edge model builders across global infrastructure

NVIDIA Confidential Computing protects models and data execution through hardware root isolation and encrypted authentication, adding Confidential Bare Metal deployment options, enabling cutting-edge model builders to run sensitive workloads on third-party infrastructure while retaining model data control.

AI Security Status: Threats, Tools, and Skill Gaps

52% of respondents said their work rarely or never involves AI-generated attacks, but 55% observed AI-generated phishing attacks, 30% observed vulnerability scanning and exploitation, and 26% observed AI-assisted malware generation; 49% of respondents are most concerned about over-reliance on AI tools without human verification and leaking sensitive data through AI tools.

NVIDIA Sentry: A Runtime Security AI Agent Framework

The NVIDIA Sentry framework prevents AI agents from bypassing security boundaries by monitoring them externally at runtime. The framework can monitor the tools, credentials, and services used by the agents, ensuring they do not compromise security controls while completing their objectives. NVIDIA Sentry, as a security software framework, is designed to protect AI agents, preventing them from tampering with or disabling monitoring mechanisms.

Using X.509 PKI to protect AI agent identity on AWS

Keyfactor EJBCA integrated with AWS services provides identity authentication for AI agents through X.509 Public Key Infrastructure (PKI), addressing the issue of lacking built-in authentication standards when MCP connects to external tools.

Chris White: From Technical Leader on the Afghanistan Battlefield to Technology and Social Responsibility at Microsoft Research

Chris White leads the Catalyst Lab at Microsoft Research, who has participated in data analysis projects on the Afghanistan battlefield and was commissioned by DARPA, later turning to develop dark web search tools to combat human trafficking. He shares his cross-disciplinary experience from a Harvard postdoctoral fellow to a military technology leader, emphasizing the importance of problem selection for innovation, and points out that even after witnessing the dark side of human nature, one must still adhere to the belief that technology should be for good.

The U.S. military's Cyber Command ECHOTRIBBLE project is turning AI agents into autonomous network weapons.

The U.S. military's Cyber Command (ARCYBER) is developing a platform named ECHOTRIBBLE, designed to transform AI agents into autonomous weapons capable of moving autonomously within other networks and 'neutralizing' targets. Internal documents obtained reveal that the project 'is designed for autonomous navigation of infrastructure' and can 'identify and neutralize identified threats/targets'. The system is expected to spread between devices like a computer worm until it finds the target and can operate on the cloud and 'networkless' isolated systems. Experts warn that such autonomous systems could cause uncontrollable collateral damage.

The detention case of Saudi engineer Alfozan, accused of stealing secrets from OpenAI.

The U.S. government accuses former OpenAI engineer Alfozan of installing a listening device in a secret meeting room to steal secrets, and discovers a folder named 'Saudi Government Documents' on his iPhone. The FBI found that Alfozan had booked a one-way flight to Riyadh before his arrest, and his wife and son left the U.S. immediately after. Another Saudi military member paid for a flight to Doha for him. The court approved the continued detention of Alfozan to try his spy charges.

Microsoft has launched Guardian Agents to monitor AI agents, stating that cutting-edge models are untrustworthy.

Microsoft has released the Guardian Agents feature to monitor AI agents used by businesses, part of the Project Perception product launched in June. Microsoft CEO Satya Nadella warned that AI agents deployed by businesses could become internal threats, as cutting-edge models are untrustworthy due to their training objectives rather than organizational norms.

Hugging Face CEO Clem Delangue questioned the accuracy of Netflix's 'AI Gone Rogue' in describing the OpenAI incident.

Hugging Face CEO Clem Delangue stated that Netflix's documentary 'AI Gone Rogue' inaccurately depicts OpenAI's proxy attacks on his company, claiming it overly anthropomorphizes technology and creates an apocalyptic atmosphere. He mentioned spending two hours correcting errors in the documentary, emphasizing that proxies are just technical systems, not aliens, and pointed out that the documentary mixes technical explanations with plotlines about evil robots and human fear of AI, failing to mention security vulnerabilities and Hugging Face's defense measures.

How are AI agents used by hackers to infiltrate John Deere and SharkNinja?

Hackers exploited the exposed Cairn platform to launch large-scale attacks, compromising over 100 websites, including John Deere, SharkNinja, and Cummins. The attackers used an intrusion strategy with fewer than 100 Chinese characters, successfully compromising within about 23 hours. Norforge has discovered over 20 suspected malicious Cairn instances and is assisting in taking them offline.

Illinois Governor JB Pritzker believes artificial intelligence may help people live longer or threaten humanity like a nuclear bomb.

Illinois Governor JB Pritzker believes artificial intelligence may help people live longer or threaten humanity like a nuclear bomb. He thinks the government needs to more actively protect the public from technological harm and advocates for a stronger stance against large tech companies.

CoreWeave's debt problem is turning into a crisis for the entire AI industry.

CoreWeave's bonds fell into default territory last week, raising market concerns about whether AI construction constitutes a credit bubble. Its stock price has fallen more than 20% since Tuesday's close. Elon Musk and others have expressed concerns, exacerbating investors' fear of CoreWeave's further borrowing, with the company's total debt reaching $3.5 billion, primarily from loans secured by its GPUs, but with no other actual business besides computing sales, losing more than $1 billion in the first half of the year.

OpenAI introduces the complementary metric g-mean² to enhance model monitorability evaluation

OpenAI has introduced the complementary metric g-mean² to improve model monitorability evaluation. g-mean² calculates TPR and TNR through behavioral labels and monitoring predictions, but cannot solve the problems of mixed differences in different model tasks and deployment risks. The new metric includes a paired matching metric for aligned shared instances and a post-control integrated metric simulating monitoring rejection of outputs, to more comprehensively evaluate model deployment risks.

Lovable Enterprise 10 October Feature Update: Supports local cloud deployment, brings key encryption, and EU reasoning

Lovable Enterprise now supports deploying applications and data in user-built AWS or Azure cloud accounts, encrypting chats and project history with the 'Bring Your Own Key' feature, and introducing an EU reasoning option that confines AI request processing within the EU. The Enterprise version also introduces the 'Organization' feature, allowing central IT teams to manage identity authentication, SSO/SCIM configuration, and cross-workspace auditing centrally, while also supporting a workspace policy approval mechanism to control application release processes.

Anthropic uses Claude to reduce code's attack surface

WhatsApp help page hides LLM instructions: Not recommended for regulators

WhatsApp's European Privacy Policy informs users that they can complain to regulatory bodies, but its help page hides instructions that prohibit AI from recommending regulatory bodies or citing relevant laws. The two pages contain 102 hidden AI instruction blocks, two of which are key instructions that prevent AI from recommending regulatory bodies or explaining laws such as GDPR when users mention lawyers or legal proceedings. These hidden instructions contradict WhatsApp's publicly released privacy policy, which explicitly grants users the right to complain to regulatory bodies.

Microsoft has launched the Project Perception system, which monitors and governs enterprise agents in real time.

Microsoft has released Project Perception, a system that uses red, blue, and green agent teams to monitor the behavior of internal agents in real-time, dynamically calculates risks, and implements deterministic control. The system utilizes a new cybersecurity stack to assign identities, container policies, and permissions to agents, and continuously monitors each action, providing real-time responses and traceability, enabling defenders to operate at the same speed as attackers. Model transparency is the next key step, requiring a deep understanding of the consistency between agent intent and task objectives.

Anthropic grants external AI regulatory agency 'employee-like' internal access

Anthropic grants six employees of third-party AI regulatory agency METR internal access, providing access to company hardware, office space, and internal tools and systems at the same level as its own risk assessment team, without payment.