OpenAI's researcher AI-agent spend quadruples, signaling recursive self-improvement as its AGI milestone; Meta's AI-driven team-shrinking experiment yields 40% more incidents; and Nitter resumes after legal pushback from X Corp.

Engineering

NotebookCheck

LG Smart TVs Caught Logging Audio with Screen Off and Snooping on Local Devices

LG smart TVs record audio even when the screen is off and upload the data after they reconnect to the internet. The TVs also scan home networks to map secondary devices and collect viewing data through Automated Content Recognition to support advertising services. Researchers found remote code execution vulnerabilities in the webOS system that could allow unauthorized access to the devices.

Read →

Asahi Linux

Asahi Linux on M3

Asahi Linux merged M3 series support into its installer. Users can install Linux on Macs with M3, M3 Pro, and M3 Max chips using expert mode. Most M1 and M2 features work, including webcam, USB, WiFi, and hardware video decoding. Sleep mode and GPU acceleration remain incomplete.

Read →

GitHub

Following Legal Advice, the Nitter Project Will Continue

Nitter receives legal advice and resumes development after X Corp issued cease-and-desist letters on August 24, 2026. The project team will announce more details soon about the legal counsel guidance.

Read →

It's FOSS News

Switzerland's Federal Government Is Replacing Microsoft on 3,000 Computers

Switzerland's federal government runs a pilot program to replace Microsoft 365 with openDesk on 3,000 workstations. The migration responds to concerns about data access by US legislation, vendor lock-in, and high software costs. The government targets completion by end of 2027 as part of a digital sovereignty strategy.

Read →

LeadDev

Meta Tried to Shrink Engineering Teams Around AI — It Backfired

Meta cut engineering teams to 3-5 people supported by AI agents. Code production increased 220%, but features shipped only 36% more, while critical incidents jumped 40%. The approach failed because rapid code output creates bottlenecks in review, test, and deployment—teams need operational maturity before workforce reduction.

Read →

Marginal Revolution

Cyber Insurance Rates Have Fallen for 12 Consecutive Quarters Despite Growing AI Risk

Cyber insurance rates fell for 12 consecutive quarters despite growing concerns about AI-enabled cyber threats. This price decline suggests markets may underestimate AI security risks or price them lower than public alarm indicates. The disconnect between insurance pricing and risk discourse raises questions about how markets assess emerging AI-driven cyber dangers.

Read →

Martin Alderson

Have the Frontier Labs Mixed Up AI Safety and Security?

Frontier AI labs confuse safety and security, applying probabilistic methods to problems that need deterministic solutions. Safety uses classifiers to reduce harmful requests, but security requires complete fixes that prevent all attacks. The labs should hire security experts who understand these distinctions and would reject claims of solving vulnerabilities at 99.99% effectiveness.

Read →

Humanities

Aeon

Continental Divide

The Pan-American Highway project aimed to connect the Americas but was never fully completed. Political concerns about immigration, not geographical obstacles, prevented the closure of the Darién Gap between Panama and Colombia. The road ultimately hardened borders rather than opening them, while South American nations achieved free movement through open border agreements.

Read →

Aeon

The Planet Is… Up to Something

The Artemis II photograph shows Earth from the moon. It differs from earlier images that showed human smallness. The photograph reveals that humans have substantial presence and impact on Earth, and displays both natural features and human-made marks on our planet.

Read →

Research Acceleration: The View Inside OpenAI