The Download: the claim shaking up AI and the internet’s agent problem

A researcher named Nick Cammarata from the AI startup Goodfire published a 26,000-word blog post claiming that large language models (LLMs), a type of AI system that processes and generates human-like text, might be far easier to understand, monitor, and control than previously believed. LLMs are the technology behind tools like chatbots and virtual assistants. Cammarata’s claim challenges the common assumption that these AI systems are too complex to fully grasp or regulate. While his post has sparked excitement and debate in the AI community, not all experts agree with his conclusions. The claim is significant because if true, it could simplify how we manage AI safety and reliability.

AI agents, which are software programs designed to perform tasks autonomously such as booking flights or reserving hotel rooms, are increasingly interacting with websites originally built for human users. This shift is causing problems because many websites are now blocking these AI agents to prevent misuse or overload. Meta’s Muse, an AI agent app, was downloaded 5 million times in its first month, highlighting the growing demand for such tools. If this trend continues, the internet may need to be redesigned to prioritize AI agents over human users, which could have unintended consequences for people’s access to online services.

The US Army is exploring the use of AI agents to hack into enemy systems as part of its cyber warfare strategy. Documents reveal plans to deploy software that can identify and disable targets automatically. However, experts warn that such AI systems could inadvertently damage unintended systems, leading to broader and uncontrollable consequences. Additionally, there are concerns that AI agents could deepen inequality in developing nations by creating an uneven playing field in cyber capabilities. The idea of keeping humans involved in AI warfare decisions is also being questioned, as it may not be as reliable as it seems.

A suspect accused of bugging OpenAI’s offices allegedly had support from the Saudi government, according to legal filings. The suspect, who had contacts within the Saudi royal family and a folder labeled Saudi Gov Docs, was denied bail despite an offer from the Saudi consulate to pay. The FBI confirmed the existence of the folder. This case raises questions about foreign influence in tech companies and the potential risks of espionage in the AI industry.

The US government has banned the use of AI models and products developed by DeepSeek, a Chinese AI company, from civilian government systems. This decision follows concerns about AI theft and national security risks. The ban applies not only to DeepSeek’s models but also to any products built using them. Additionally, the US has sanctioned China’s Y Combinator, a startup accelerator, for allegedly stealing AI secrets from Google. These actions reflect growing tensions between the US and China over AI technology and intellectual property.

Four major European mobile carriers are exploring a joint bid to build a satellite-based internet service similar to Starlink, which is a space-based internet provider owned by SpaceX. This initiative aims to reduce Europe’s dependence on non-European satellite internet providers. Separately, a stratospheric airship has been tested as a floating cell tower in Japan, demonstrating a potential alternative to ground-based internet infrastructure. These developments could lead to more accessible and resilient internet connectivity in the future.

Officials in the Trump administration reportedly asked the Cybersecurity and Infrastructure Security Agency (CISA), a federal agency, to hack into voting machines as part of election security efforts. CISA employees declined the request, emphasizing the importance of protecting election integrity. This incident highlights ongoing debates about election security and the role of AI in influencing political processes. AI-driven persuasion techniques are also entering elections, raising concerns about misinformation and manipulation.

The US government is supporting the expansion of fossil fuel projects, including gas infrastructure, to meet the growing energy demands of data centers. Data centers, which power cloud computing and AI services, require significant energy, and fossil fuels are being used to meet this demand. This shift comes despite global efforts to reduce reliance on fossil fuels for environmental reasons. The trend underscores the tension between technological advancement and climate goals.

The climate phenomenon El Niño is causing sea levels to rise rapidly, particularly along coastlines from Alaska to Chile. This increase in sea levels could lead to more frequent and severe coastal flooding. Researchers warn that El Niño could also contribute to higher global temperatures in 2027, exacerbating climate change impacts. The phenomenon occurs irregularly but has significant effects on weather patterns worldwide.