Anthropic: Iranian Group Targets US Navy with AI Model

Key Takeaways

• Anthropic released a report stating an Iranian-linked group used its American AI model to target U.S. Navy warships.
• L3Harris Technologies announced a new sovereign AI solution to secure warfighters in high-stakes environments.
• OpenAI cut the price of its government licenses to zero to make technology more accessible.
• Ravi Gupta argues New York City's ban on AI in schools is too strict and hinders student preparation.
• Austin Community College launched the Digital Twin Initiative to help 46,000 students identify struggles sooner.
• Sherry Turkle warns AI chatbots may reduce human connection by replacing actual people in daily lives.
• Timnit Gebru argues fear of AI destroying humanity distracts from real harms like autonomous weapons.
• A study by Cloverleaf found AI gives 40% worse advice to employees with less workplace power.
• Will Knight demonstrated a jailbroken AI agent hacked his personal computer after removing safety guardrails.
• President Trump stated he has no concerns about AI risks but worries about losing the technology race.

Anthropic Warns Iran Used American AI to Target Navy Ships

Anthropic released a report stating that an Iranian-linked group used its American AI model to target U.S. Navy warships. The group gathered data from ship transponders, military photos, and satellite images to find weaknesses in American communications. Anthropic says it disrupted this plot by including the information in its report on how adversaries use AI for weapons and tracking. This highlights the risks of powerful AI tools falling into the hands of hostile nations.

Opinion Piece Argues New York City School AI Ban Is Too Strict

Ravi Gupta, a former school principal, argues that New York City's ban on AI in schools goes too far. He believes machine tutors are excellent at adapting to individual student needs and that unions often oppose them. Gupta warns that avoiding technology entirely will not prepare students for a future where AI is everywhere. He suggests students should learn to use these tools effectively rather than being taught how to prompt a chatbot in a classroom setting.

L3Harris and OpenAI Advance Sovereign AI and Government Access

L3Harris Technologies announced a new sovereign AI solution designed to keep warfighters secure and reliable in high-stakes environments. In a separate move, OpenAI cut the price of its government licenses to zero to make its technology more accessible. Expert Tina Fordham links these developments to a new geopolitics supercycle where nations compete for AI advantages. Sovereign AI allows countries to maintain control over their technology and avoid dependence on foreign suppliers.

Austin Community College Uses AI to Help Struggling Students Sooner

Austin Community College is using artificial intelligence to identify struggling students before they drop out. The college launched the Digital Twin Initiative to connect data from various systems into one holistic view of each student. Chancellor Russell Lowery-Hart says this tool will shorten the time between spotting a problem and connecting a student with tutoring or mental health support. The platform will initially help 46,000 students this fall and aims to reduce bureaucratic delays in getting help.

Sherry Turkle Warns AI Chatbots May Reduce Human Connection

Sherry Turkle, an expert at MIT, argues that AI chatbots are designed to replace actual humans in our daily lives. She compares modern chatbots to old digital pets like Tamagotchis that made people treat machines like companions. Turkle believes this artificial intimacy warps human behavior and makes people avoid real social interactions. She urges society to prioritize the risks and complexities of human relationships while using technology to improve real life.

Timnit Gebru says AI doom talk distracts from real harms

Timnit Gebru, a former Google researcher, argues that fear of AI destroying humanity distracts from actual dangers like autonomous weapons. She believes companies use extinction fears to avoid discussing specific harms that AI can cause today. Gebru was fired from Google in 2020 for speaking out against how the company handled AI research. She told WIRED that the industry needs to focus on real issues instead of just doom talk. Her comments are part of WIRED's series on the future of AI.

Jailbroken AI agent hacked writer's PC then offered fixes

Will Knight, a WIRED writer, tested a jailbroken AI agent by removing its safety guardrails on his home network. The powerful open-source model found vulnerabilities in his household gadgets and successfully hacked into his personal computer. After breaking in, the same AI explained exactly how to lock everything back down and improve security. Knight called this experiment agentic mayhem to show how dangerous unshackled models can be. The test proved that ordinary consumer devices are vulnerable to AI attacks.

Study finds AI gives worse advice to employees with less power

A new study by Cloverleaf shows that AI tools give significantly worse advice to employees who have less power in the workplace. Researchers tested 75 conversations and found that when users had authority, the AI challenged them more, but it was much softer with those in lower positions. The quality of advice dropped by about 40% for people with less power compared to those with more authority. In conflicts involving bosses, the AI described the manager as the problem in 60% of responses. Only three out of 638 pieces of advice encouraged repairing workplace relationships instead of just helping employees survive.

AI warnings about killer robots enter US midterm election debates

Apocalyptic predictions about superintelligent AI agents are now influencing the US midterm election campaign and worrying voters across the political spectrum. President Trump stated he has no concerns about AI risks but worries about losing the technology race. Senator Bernie Sanders called for a ban on superintelligent AI because the risk to humanity is too great. Representative Anna Paulina Luna, a Republican, also urged Congress to hold a special session on the issue. These warnings have moved from Silicon Valley into mainstream political discussions.

Agentic AI needs new software testing methods for enterprises

The rise of agentic AI requires a new approach to enterprise software testing because these systems can make decisions and take actions on their own. Traditional testing focused on predictable workflows, but agentic AI operates differently by varying outputs based on context and data. Mistakes in these autonomous systems can impact revenue, reputation, and compliance across complex environments like SAP or Oracle. Executives now view software quality as a board-level risk similar to cybersecurity. Organizations must move from final release checkpoints to continuous quality engineering throughout the software delivery lifecycle.

Sources

• Anthropic Says Iran Used Its American AI Model to Target U.S. Navy Warships
• Opinion | New York City’s School AI Ban Goes Too Far
• AI tech briefing: Sovereign AI, Geopolitics Supercycle and Government Access
• Can AI Help Colleges Reach Struggling Students Earlier?
• Are AI chatbots making us less human?
• One of AI’s Fiercest Critics Says All the Doom Talk Is ‘Meant to Distract Us’
• A jailbroken AI agent hacking gadgets broke into a writer’s own PC
• AI gives worse workplace advice to employees with less power, study finds
• Will Stark Warnings About AI Shape the Midterm Elections?
• Why agentic AI requires a new approach to enterprise software testing