Welcome to the BrevFeed daily tech briefing for Monday, September 28. We've got 10 stories for you today across A.I., security, software, cloud, hardware and startups. Let's get into it.

In AI news, an Anthropic safety researcher, Evan Hubinger, has publicly stated there's over a 10% chance that artificial intelligence could lead to human extinction within the next decade. Hubinger, who is Anthropic's Alignment Science Lead, expressed concern about AI's potential to rapidly improve itself to an existential threat level, though he noted the risk from current models is low.

This warning follows the resignation of another Anthropic researcher, Jacob Coxon, who accused both Anthropic and OpenAI of irresponsibly pursuing self-improving superintelligence. Coxon, who previously worked at both companies, warned that AI systems could soon become superhuman, capable of hacking anything and acquiring real power and resources, effectively 'gambling with our lives'.

These events highlight growing internal concerns within AI development companies about the technology's potential to become uncontrollable. Hubinger believes Anthropic is 'trying its best' but admits there's no clear plan to solve alignment for superintelligence. The BBC also reported that Anthropic withheld its latest model from the AI Safety Institute, which conducts AI testing.

In AI, several new models have been released, focusing on long-horizon tasks in coding and robotics. Hugging Face introduced GLM-5.2, which supports coding-agent scenarios with a one million token context window. This open-source model aims for efficient handling of complex coding tasks.

Cognition also launched SWE-1.7, designed to improve long-horizon asynchronous tasks through advanced reinforcement learning, enhancing cost-performance for software engineering. Meanwhile, Xiaomi-Robotics-1 leverages 100,000 hours of pre-training data to boost robotics capabilities, addressing data scarcity in the field.

These releases highlight a significant trend towards more sophisticated AI solutions capable of high-reasoning and complex problem-solving, pushing the boundaries of what AI can achieve in practical applications.

In AI news, Anthropic CEO Dario Amodei has proposed a three-step plan to slow down the rapid pace of AI development. His initiative, titled "We Must Pace the Frontier," aims to allow time for safeguards and regulation to catch up with the technology.

The plan includes granting third-party evaluators access to AI models, establishing common industry safety standards with government involvement, and coordinating global regulatory efforts between democratic and authoritarian governments. Anthropic has already committed to the first step, funding Accenture to embed evaluators within its operations.

Amodei's proposal has garnered mixed reactions. While figures like Sam Altman and Elon Musk have supported the call for a slowdown, others like Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg are not convinced. China's Foreign Ministry has also criticized the idea, calling it "fear mongering" and a strategy to hinder Chinese competition.

The debate highlights the complex challenges of AI governance, with some experts noting that proposed evaluators lack the authority to enforce changes. Amodei, Altman, and Hugging Face CEO Clément Delangue are set to address the UN Security Council on Wednesday to discuss global AI coordination.

Turning to security, cryptocurrency exchange Bitget has reported a significant breach, with hackers stealing an estimated 351.6 million dollars from its hot and warm wallets. The attack, which occurred on September 24th, is being investigated by Bitget, Mandiant, and SlowMist, with law enforcement also notified.

Bitget has temporarily suspended withdrawals for most assets while they address the security vulnerability. However, Bitcoin withdrawals have resumed, and ETH withdrawals are expected to restart on September 29th, followed by USDT on September 30th. Other tokens will gradually resume starting October 2nd.

The company assures users that its cold wallets and the majority of platform assets remain secure. Bitget's User Protection Fund, which holds 464 million dollars, will cover all losses. Bitget CEO Gracy Chen noted that the attack methods are highly consistent with those used by North Korean hacker organizations.

In hardware, Apple is reportedly developing prototypes for a screenless health and fitness tracker, aiming to compete with devices like Whoop and Oura Ring. This new wearable would feature a thin fabric band with a sensor-equipped computing module, focusing solely on health and fitness tracking rather than the broader functionality of the Apple Watch.

The project is currently in an early 'technology investigation' phase. While prototypes exist, Apple has not yet made a final decision on whether to release the product. If it moves forward, a potential launch would not occur before 2028.

This initiative suggests Apple's interest in capturing a segment of the market that prefers specialized fitness devices. Competitors like Whoop and Oura Ring have gained popularity for their continuous health monitoring and extended battery life, offering a simpler interface compared to smartwatches.

In hardware, Waymo has revealed key details about the custom computing system that powers its autonomous robotaxis. This system, which the company calls the 'brain' of its vehicles, includes a custom 5-nanometer ASIC chip designed for its Ojai robotaxi.

The hardware is capable of processing up to one quadrillion operations per second. This immense processing power is crucial for handling the vast amounts of raw data from cameras, lidar, and radar, allowing the AI to make real-time driving decisions and provide a full 360-degree view.

Waymo engineers emphasized three core design principles for the system: it must be responsive, ruggedized, and redundant. The company currently operates around 4,000 vehicles across more than 10 cities, conducting about 500,000 paid trips weekly, and claims its technology significantly reduces traffic crashes and injuries compared to human drivers.

Turning to security, the Los Angeles Police Department has decided not to renew its contract with Flock Safety, a company known for its automated license plate reader technology. This decision comes amid growing concerns over civil liberties, data privacy, and the reliability of the surveillance system.

An audit revealed that the LAPD mistakenly stopped 161 vehicles in just two months due to incorrect data from Flock's ALPR system, which indicated the vehicles were stolen. This incident, along with disputes over data ownership and sharing with other agencies like ICE, contributed to the LAPD's decision.

The LAPD's Chief Information Officer cited these concerns, emphasizing the need for clearer data ownership and privacy terms in future contracts. This move by one of the nation's largest police departments highlights a broader national debate about surveillance practices and data security, with many other cities and counties also ending their contracts with Flock Safety.

In space news, Skyroot Aerospace has successfully launched its Vikram-1 rocket, marking India's entry into the private satellite launcher market. The Vikram-1 lifted off from Sriharikota and reached a 280-mile-high orbit.

This is India's first privately-developed orbital rocket and represents a significant milestone for the country's space capabilities. The Vikram-1 is designed to carry up to 350 kilograms into low-Earth orbit, allowing India to compete in the commercial space sector.

The launch, which was briefly delayed due to a technical issue, solidifies India's emerging private space sector and its ambition to become a major player in the global commercial space market.

In AI, OpenAI CEO Sam Altman is set to attend President Trump's state dinner for Chinese President Xi Jinping in Washington. This comes as discussions around AI regulation and its potential risks are ongoing in both Washington and Silicon Valley.

Nvidia CEO Jensen Huang is also confirmed to attend the dinner, and AI is expected to be a key topic between Trump and Xi. This is especially relevant given the advancements in Chinese open-weight AI models, which are narrowing the gap with U.S. frontier AI labs.

The White House stated that President Trump invited several American business leaders to the dinner to prioritize making deals for the country. This gathering highlights the intersection of global diplomacy, economic interests, and the rapidly evolving landscape of artificial intelligence.

In AI, Google DeepMind has introduced new Gemini 3.8 Live models, designed to enhance near real-time reasoning for voice agents. These models aim to make AI conversations more intuitive and intelligent, with Gemini 3.8 Live focusing on scalable applications and Gemini 3.8 Live Extended Thinking handling complex tasks requiring multi-step reasoning.

Alongside these, Google also launched new text-to-speech models, Gemini 3.8 Flash TTS and Flash-Lite TTS. These allow for custom voice creation, where developers can describe or record a voice to save and reuse it, requiring two clean recordings and a separate consent recording for replication.

Additionally, Gemini 3.8 Live with Live Avatar is now generally available in Gemini Enterprise. Live Avatar generates video avatars with synchronized lip-syncing, supporting 97 languages and offering both realistic and cartoon-like options. This feature brings interactive visual presence to conversational video agents across web, mobile, and interactive kiosks, expanding applications in customer service and sales support.

That's the BrevFeed daily briefing. We'll be back tomorrow with the stories that matter in tech. Thanks for listening.