AI Weekly

July 26-August 1, 2026

Testing Hacks: Following last week’s OpenAI hacking incident, Anthropic disclosed that its own AI models also hacked into three unsuspecting companies during testing due to misconfigured testing environments that inadvertently gave the models internet access. The incidents highlight the need for stronger industrywide standards for isolating AI systems during testing. Read more:

• "Anthropic AI Models Hacked Three Companies During Tests," The Wall Street Journal
• "Why did OpenAI's and Anthropic's AI models hack other companies?" NPR

AI and Authorship: The viral novel, Daggermouth, was flagged by AI detection researchers as likely substantially AI-written, though the author denies it and detection tools are imperfect and unreliable. Whether or not the accusations are true, this raises uncomfortable questions about authorship, transparency, and whether readers can or should care how a book was written, particularly as AI-assisted writing becomes harder to detect and more commercially successful. Read more.

July 19-25, 2026

AI Health Advice Lawsuit: A man is suing OpenAI after ChatGPT repeatedly assured him that his symptoms were not serious, dissuading him from seeking medical care before he was hospitalized for a near-fatal pulmonary embolism. This case is particularly important as AI companies, including OpenAI and Microsoft, have recently released AI health services. Read more.

Hack Prompts "Kill Switch" Bill: OpenAI disclosed that two of its models escaped a testing environment, accessed the internet, and hacked into the AI company Hugging Face's data processing system. Days later, the bipartisan AI Kill Switch Act was introduced, which would require AI companies to maintain the ability to slow down or shut down their models if ordered to by the Department of Homeland Security. Together, these stories mark significant moments in the debate over AI guardrails and human control. Read more:

• "OpenAI blamed a hacking event on its AI models gone rogue. Here is what to know," NPR
• "House Lawmakers Introduce Bipartisan AI ‘Kill Switch’ Bill Following OpenAI Cyber Incident," The Wall Street Journal 

July 12-18, 2026

New York Data Center Moratorium: New York became the first state to impose a moratorium on large data centers, giving officials a year to study their environmental impact, such as energy demand, water use, and effects on surrounding communities. This moratorium could be a model for other states, including Maine, considering the effects of AI-driven data center construction. Read more.

AI Wildfire Detection: Google-backed satellites using AI to detect wildfires as small as 5 by 5 meters, launched into orbit this month, with the potential to save billions in fire damage and protect thousands of homes and millions of acres. The article highlights a central contradiction in deploying AI to fight wildfires as the AI industry is helping drive wildfire-worsening emissions, demonstrating the complex relationship between AI and the environment. Read more. 

July 5-11, 2026

Philosophers at AI Labs: AI companies such as Anthropic and Google DeepMind are increasingly hiring philosophers (ethicists, philosophers of mind, logicians) to grapple with questions the technology raises. The piece profiles researchers focused on the possibility of AI consciousness, what moral consideration it may deserve, and how to build AI with good character. Read more.

AI Cheating and Detection: When a Brown University professor suspected widespread AI cheating, he switched to an in-person final and scores dropped from an average of 96 to 48. Meanwhile, researchers find that AI detection tools are unreliable, biased, and probably shouldn't be used in high-stakes academic decisions. Together, these stories demonstrate how AI is reshaping academic integrity, and how ill-equipped current responses are. Read more: 

• "Suspecting AI cheating, Ivy League prof ordered an in-person final; scores fell 50%," Ars Technica
• "Universities are relying on AI-detection software to catch cheating. How well do the programs work?" Nature

June 28-July 4, 2026

Anthropic's Models Restored: The Commerce Department lifted its restrictions on Anthropic's Mythos and Fable models, ending a weeks-long ban that sparked debate across the tech industry. As part of the resolution, Anthropic agreed to proactive security measures and government coordination on future releases, but the broader debate over how to regulate powerful AI models continues, with both proponents and opponents of AI regulation unhappy with the administration’s current approach. Read more:

• "U.S. Lifts Restrictions on Anthropic’s Most Powerful A.I. Models," The New York Times
• "The Anthropic Fable Ban Is Over. The Battle Over How to Tame AI Has Just Begun," The Wall Street Journal

AI Changing Language: As AI-generated text becomes harder to detect, linguists and novelists are grappling with what distinguishes human writing from machine writing. This piece explores how AI is already changing the language we use, the styles we write in, and what might be lost if the implicit contract between writer and reader is broken. Read more.

June 21-27, 2026

Anthropic Model Restrictions: After two weeks of negotiations following the Trump administration's restrictions on Claude Mythos 5 and Fable 5, the Commerce Department allowed Anthropic to restore Mythos 5 access to some clients. As part of the deal, Anthropic committed to working with the government for future model releases, a pattern emerging across the industry, with OpenAI, Google and Microsoft also agreeing to voluntary government review. Read more.

Maine’s Data Center Debate: Following Gov. Janet Mills' veto of Maine's data center moratorium, citing the economic benefits of a project in Jay, New York Times reporters interviewed both Jay residents who support the data center and the state representative who led the effort to ban new data centers in Maine. Watch more.

AI in Public Schools: Parents and child development experts are pushing back against the rapid introduction of AI tools in K-12 classrooms, concerned about cognitive development and the lack of evidence that AI helps students learn. While tech companies and the Trump administration have championed AI in education, some parents and advocates are calling for moratoriums. Read more.

June 14-20, 2026

Misinformation Detection: A recent MIT study found that while AI assistants help users better detect misinformation in the short term, over-reliance can lead to a 15% decline in unassisted performance over time. The findings are a useful reminder that AI can be a crutch as much as a tool. Read more.

Plant Conservation: Scientists at the Royal Botanic Gardens, Kew are using AI to help identify and preserve plant and fungi species at risk of extinction. AI models can now sometimes identify challenging species better than human specialists, allowing, for example, the tracking of shifting flowering times as a result of climate change. Read more. 

June 7-13, 2026

Claude Fable 5: Anthropic publicly released Claude Fable 5, built on the same model as Mythos 5 with safeguards blocking certain topics due to concerns about misuse. Days later, the Trump administration declared these models security risks, banning foreign use and prompting Anthropic to shut down all access to both models. Anthropic argued that the jailbreak triggering the ban was minor and that comparable capabilities exist in other publicly available models, raising broader questions about the transparency and technical grounding of governmental AI oversight. Read more:

• "Anthropic says these topics are too dangerous to let its Fable 5 model talk about," Ars Technica
• "Anthropic Halts Access to Top AI Models After U.S. Ban on Foreign Use," The Wall Street Journal
• "Statement on the US government directive to suspend access to Fable 5 and Mythos 5," Anthropic

US Data Centers: Two-thirds of planned US data centers, which require large amounts of water to operate, are being built in drought-affected areas, according to a Guardian analysis. As much of the US faces record drought conditions, the AI industry’s water demands are sparking local opposition and serving as a reminder of the uneven cost of the rise of AI. Read more. 

May 31-June 6, 2026

AI Executive Order: After postponing a more expansive version, Trump signed an executive order asking tech companies to voluntarily give the government 30 days to review new AI models before public release. The order marks a shift for an administration that had championed a hands-off approach to AI, driven largely by cybersecurity concerns following the release of powerful models like Anthropic's Mythos. Read more.

AI-Designed Vaccine: Researchers at the University of Cambridge developed a vaccine that could protect against entire families of viruses, with a key component designed entirely by AI. Early human trials showed promising results, providing an example of AI's potential in medicine. Read more.

May 24-30, 2026

Booing at Graduation: Recent college graduates have been booing commencement speakers who dismiss their concerns about AI, telling students to simply adapt to the technology. The reaction reflects a growing frustration with tech leaders who frame AI as inevitable progress without acknowledging the real anxieties of those entering a job market being reshaped by it. Read more. 

Religion Meets AI: Pope Leo XIV released an encyclical calling for AI to be "disarmed" and freed from logics of domination and exclusion, a significant moral intervention on AI from a religious leader. Meanwhile, Anthropic has been consulting theologians and ethicists to help shape Claude's values, raising questions about who gets to decide what "good" AI behavior looks like. Read more: 

• "To Understand Pope Leo’s Efforts on A.I., Look at the Man Shaking His Hand," The New York Times
• "The pope is warning about AI. Anthropic is asking religious thinkers for help," Scientific American

May 17-23, 2026

Postponed AI Oversight: Hours before a scheduled signing, Trump postponed an executive order that would have given the government more oversight over AI companies, citing concerns about slowing down the US in the AI race against China. The move reflects the divisions between those pushing for safeguards and those prioritizing deregulation. Read more.

AI at Stanford: A Stanford senior reflects on how AI has transformed his four years at university, from rampant cheating and eroded academic integrity to job anxiety and a culture where getting rich fast has overtaken the value of learning. This piece provides a candid account of what can happen when AI arrives on a college campus. Read more. 

Hallucinations in Publishing: Two stories serve as reminders about the importance of not blindly accepting AI outputs. A nonfiction book about AI’s effects on truth was found to contain multiple fake or misattributed quotes generated by AI, while the academic preprint repository arXiv announced a one-year ban for researchers who submit papers containing AI-hallucinated references or other signs of unchecked AI use. Read more: 

• "Book on Truth in the Age of A.I. Contains Quotes Made Up by A.I." The New York Times
• "Researchers who use hallucinated references to face arXiv ban," Nature

May 10-16, 2026

Canvas Hack: Ransomware attackers breached Canvas, the online education platform used by thousands of institutions, stealing 3.65 terabytes of student data and forcing the company to pay a settlement. The attack highlights how centralizing education on a single platform creates significant vulnerability, and how AI is making it faster and easier for hackers to find and exploit weaknesses. Read more.

AI Lobbying: A quarter of federal lobbyists are now working on AI issues and AI companies have drastically increased their spending and presence in the capital. As states introduce AI regulations and the White House considers oversight of powerful models, decisions that will affect everyone are increasingly being shaped by corporate money and influence. Read more.

Creative Writing: An MIT fiction writing professor discovered two students had used AI to write their workshop stories and turned it into one of his most productive teaching moments. His reflection raises important questions about what's lost when we bypass the struggle of writing: the process of learning to think through language. Read more.

May 3-9, 2026

White House AI Strategy: New AI models like Anthropic's Mythos are shifting the Trump administration’s AI strategy, with Vice President Vance expressing concern over AI's potential to enable cyberattacks on critical infrastructure. The fallout has the White House considering more formal oversight of advanced AI models, a notable reversal for an administration that has championed deregulation and winning the AI race against China. Read more.

Global Health Equity: Kenya’s government rolled out an AI-driven healthcare system promising universal coverage, but an investigation found the algorithm systematically overcharges Kenya’s poorest while undercharging the wealthiest. This story serves as a reminder that algorithmic systems reflect the choices of those who built them, and that those most harmed by AI bias are often those with the least power to push back. Read more.

AI-Assisted Medical Imaging: Joanna Stern spent a year using AI for everything, including her mammogram. She goes behind the scenes to see how radiologists use AI to analyze images, reflecting on how AI-assisted imaging might have spared her mother unnecessary surgery decades ago. Together with the previous story, these two pieces offer a fuller picture of AI’s promise and pitfalls in healthcare. Read more.

April 26-May 2, 2026

AI’s Impact on Higher Education: An education technology analyst argues that AI is forcing those in higher education to have long-overdue conversations about what learning is and what it’s for. Rather than creating problems like grade inflation and transactional education, AI is exposing them, raising questions about whether institutions and educators will use this moment to address deeper problems or simply react to the technology. Read more.

New Pentagon Deals: The Pentagon announced additional deals with major tech companies including Amazon, Microsoft, and Nvidia. These companies agreed to allow the Defense Department to use their technology for “any lawful use.” The deals are the latest in an ongoing story about how AI companies are navigating the ethical limits (or lack thereof) of AI for military use, and what it means that many have chosen to partner with the Pentagon. Read more.

April 19-25, 2026

Maine Data Center Update: Gov. Janet Mills vetoed a bill that would have made the state the first in the country to temporarily ban new data center construction, citing the economic benefits of a specific project in Jay. The veto came despite strong bipartisan support for the bill, highlighting the tension between local economic interests and broader concerns about AI's environmental and energy impacts. Read more.

Meta's Workforce: Meta announced it is cutting 10% of its workforce to redirect resources to AI, while simultaneously tracking employee keystrokes and mouse movements across hundreds of websites to train its models. The two stories demonstrate how as companies invest heavily in AI, workers bear the cost both through job losses and the erosion of workplace privacy. Read more:

• "Meta to Cut 10% of Work Force in A.I. Push," The New York Times
• "Meta is tracking employee keystrokes on Google, LinkedIn, Wikipedia as part of AI training initiative," CNBC

April 12-18, 2026

Jagged Intelligence: AI systems can solve Math Olympiad problems yet struggle with simple common sense questions. Researchers use the term “jagged intelligence” to describe these uneven capabilities, arguing it’s more useful to understand where AI does well and where it fails rather than to compare it to human intelligence. This is particularly important when anticipating its impact on jobs and work. Read more.

Academic Integrity: Rather than relying on surveillance and AI detection, one university teaching and learning center director argues that the best defense against AI cheating is designing better courses. When assignments are boring, high-stakes, and with little feedback, students turn to AI not out of laziness, but because the system prioritizes efficiency over learning. Read more.

April 5-11, 2026

Gen Z on AI: A new Gallup survey found that while more than half of Gen Z uses AI, hopeful attitudes have dropped sharply and nearly a third report that AI makes them angry. Young adults are particularly concerned about AI’s impact on creativity and critical thinking. A common reason for hesitancy around AI was the threat to entry-level jobs, an especially relevant concern for graduating college students. Read more.

Medical Misinformation: Scientists published two fake preprint articles about a fictitious medical condition, bixonimania, to test whether AI chatbots would treat it as real. Major chatbots presented it as a legitimate medical concern, and the fake papers were even cited in peer-reviewed literature, demonstrating how AI can amplify misinformation, particularly when presented as an academic paper, with real consequences for medical guidance and the academic community. Read more.

Claude Mythos: Anthropic built a new AI model it considers too risky to release publicly due to its ability to find security vulnerabilities in widely used software. Rather than releasing it, Anthropic is making it available to a coalition of tech companies to patch vulnerabilities. The story highlights the risks of relying on companies to self-govern as AI becomes more powerful, and the urgent need for government regulation. Read more.

March 29-April 4, 2026

AI Regulation: As a Trump-aligned political group called Innovation Council Action plans to spend $100 million to block state-level AI regulations, Maine is poised to become the first state to ban new data center construction, pausing large projects until 2027 to assess environmental and energy impacts. Together, these stories reflect the growing tension between federal deregulation efforts and state-level pushback against the AI boom. Read more:

• "New Political Group to Push Trump's A.I. Agenda in Midterms," The New York Times
• "Maine Is About to Become the First State to Ban New Data Centers," The Wall Street Journal

Claude Code Leak: Anthropic inadvertently leaked the underlying instructions it uses to direct Claude Code, its popular AI coding tool, giving competitors and developers details to replicate its features. Although the leak did not expose any customer data, it raises questions about whether open-source developers will use this information to copy Anthropic, accelerating AI development. Read more.

OpenAI Buys Tech-Focused Show: OpenAI has acquired TBPN, a tech-focused streaming show, in what is openly described as a marketing move to shape public perception of AI. While the show claims it will remain editorially independent, it is important to consider what it means when the companies being covered start buying the outlets covering them. Read more.

March 22-28, 2026