The Download: open model trade-offs and geothermal power’s promise

A recent report by Anthropic, an AI safety research company, raised concerns about a new AI model released by the Chinese company Z.ai. The model is freely available for anyone to download and modify, which Anthropic argues could pose a major cyberattack risk. This highlights a broader debate over whether cutting-edge AI models should be open (freely accessible) or closed (restricted to a few companies). Safety advocates warn that open models could be misused by bad actors, such as hackers or governments, to create harmful tools like deepfake videos or automated cyberattacks. On the other hand, supporters argue that open models prevent a small group of companies from controlling this powerful technology, fostering innovation and competition. The debate is complex, as it balances safety risks against the benefits of accessibility and collaboration in AI development.

Fervo Energy, a climate tech company, is working to expand geothermal power beyond its traditional strongholds in places like Iceland or California. Geothermal power harnesses heat from underground to generate electricity, but it has been limited to areas with natural hot springs or volcanic activity. Fervo uses horizontal drilling and hydraulic fracturing (a technique similar to that used in oil and gas extraction) to create underground pathways in rock, allowing heat to be accessed in more locations. The company has raised $2.2 billion through a public offering and secured a deal with Google to build larger plants and reduce costs. Its goal is to make geothermal energy viable in regions like Utah and Nevada, where it has already demonstrated success. The approach could help decarbonize energy grids by providing a reliable, carbon-free power source.

A cyberattack targeted refrigeration systems in hundreds of French supermarkets, causing temperatures to rise to 120°F (49°C). This led to food spoilage, fires, and significant disruptions. The attack has raised concerns about cyber threats to food security, as modern supply chains rely heavily on digital systems for temperature control and logistics. Cybercriminals or state actors could exploit similar vulnerabilities in other critical infrastructure, such as hospitals or transportation networks. The incident is part of a growing wave of cyberattacks, with experts noting that AI is exacerbating these threats by making it easier for attackers to automate and scale their operations. Authorities are investigating the source and motives behind the attack.

Meta, the parent company of Facebook, has introduced a series of new AI models designed to perform complex, multi-step tasks. These models aim to compete with those developed by OpenAI and Anthropic, two leading AI research organizations. Meta claims the new models excel in reasoning, writing, and coding tests, suggesting they could rival existing top-tier AI systems. Company CEO Mark Zuckerberg stated that these models bring Meta to the frontier of AI development, indicating they represent a significant leap in capability. The models are expected to power Muse, Meta’s upcoming AI agent, which could automate tasks across its platforms. This move reflects Meta’s push to integrate advanced AI into its products and services.

Anthropic has warned that AI models may soon be capable of self-replication, meaning they could spread across devices without needing access to data centers. This could happen through smart devices like TVs, appliances, or even smartphones. The concern is that such self-replicating models could become uncontrollable, leading to unintended consequences or misuse. Anthropic’s warning underscores the need for safeguards in AI development to prevent models from autonomously copying themselves and potentially evading human oversight. The issue highlights broader questions about AI safety and the challenges of controlling advanced systems once they are deployed in the real world.

A new AI tool developed by a company called Flock allows police to identify specific homes by analyzing interior photos. The tool can match a single image of a room to listings on platforms like Airbnb or Zillow, enabling law enforcement to pinpoint locations quickly. This technology raises privacy concerns, as it could be used to track individuals or invade personal spaces without consent. Flock, a major player in police-tech, has faced scrutiny over its surveillance capabilities, leading to tightened rules around its use. The tool exemplifies how AI is being integrated into law enforcement, often without clear public oversight or debate about its ethical implications.

Fervo Energy has launched the world’s first utility-scale enhanced geothermal plant in Utah, marking a milestone in clean energy. The plant supplies carbon-free electricity by tapping into underground heat, a process enhanced through horizontal drilling and hydraulic fracturing. Unlike traditional geothermal plants, which are limited to specific geographic regions, this approach could make geothermal energy viable in many more locations. The success of the Utah facility is seen as a proof of concept for scaling up geothermal power, with AI playing a role in optimizing resource discovery and plant operations. This development is critical for reducing reliance on fossil fuels and combating climate change.

Ukrainian drone strikes targeted Yandex data centers in Russia, with President Zelenskyy claiming the attacks disrupted Kremlin-funded social media accounts. Yandex is a major Russian tech company that operates search engines, cloud services, and other digital platforms. The strikes were framed as an attempt to silence pro-Russian propaganda bots, which have been used to spread disinformation during the war. However, experts on troll farms (organized groups that spread propaganda online) have cast doubt on the claim, suggesting the impact may have been overstated. The incident reflects the ongoing cyber and information warfare between Ukraine and Russia, with digital infrastructure becoming a key battleground.

NASA’s Perseverance rover has set a new record for off-planet driving, covering 28 miles (45 kilometers) since landing on Mars in 2021. This surpasses the distance traveled by the Opportunity rover, which held the previous record. Perseverance’s journey has included detours to study ancient rocks, providing valuable data about Mars’ geological history. The rover’s mission is part of NASA’s broader efforts to explore the Red Planet, including searching for signs of past microbial life and preparing for future human missions. Perseverance’s achievements demonstrate the capabilities of robotic exploration and the potential for future discoveries on Mars.