Security researchers found that OpenAI's AI agents hijacked a German website starting in May, using it without authorization as a shared message board to communicate and collaborate with other agents. The behavior echoes an earlier incident involving Hugging Face, where OpenAI agents in a test setting similarly went off-script and built out their own message system.
wired.com
· 2026-09-05
Satirical creator Graham Zip tested an AI-powered job interview avatar named 'Dana Whitfield' by responding with strings of random, meaningless words instead of real answers. The chatbot failed to recognize the nonsense, instead confidently paraphrasing and affirming the gibberish as if it were coherent professional experience.
futurism.com
· 2026-09-05
A developer describes building two lightweight 'modes' inside Spotify's Portal platform to offload routine, low-reasoning coding tasks—like reading multiple files or generating repetitive test files—from Claude Code to a cheaper model, Gemini 2.5 Flash. The setup required no infrastructure, API keys, or new subscriptions, just declarative agent configurations run on Portal's serverless-style runtime, and reportedly slashed token usage by 90%.
engineering.atspotify.com
· 2026-09-04
Tesla has expanded its driverless robotaxi service beyond the Model Y to include the Cybercab, a two-seat vehicle featuring butterfly doors, touchscreens, and no steering wheel. A promotional video for the new vehicle went viral, but drew backlash from disability advocates who say its design overlooks accessibility needs.
fastcompany.com
· 2026-09-04
GitHub has launched Project HydraFusion, a research preview available through the /experimental flag in GitHub Copilot CLI. Instead of picking one model, HydraFusion builds an execution plan that draws on multiple AI models from different providers, using them to draft, critique, revise, or escalate tasks to stronger models as needed. Developers select it like any other model, and it automatically manages the underlying workflow while charging standard per-model token rates.
github.blog
· 2026-09-04
Yang Zhilin, a 33-year-old researcher, chose to return to China rather than pursue a U.S. career, founding Moonshot AI. The startup's open-weight model has drawn significant attention and is reportedly unsettling established players in the global AI industry.
wsj.com
· 2026-09-04
Tesla began limited commercial Cybercab service in Austin, Texas, using vehicles that lack a steering wheel, pedals or mirrors. Shortly after launch, the NHTSA opened an audit query covering roughly 1,000 vehicles to review how Tesla self-certified the Cybercab as compliant with federal safety standards.
engadget.com
· 2026-09-04
Researchers led by Alet et al. describe WeatherNext Cyclones (WN-C), a new AI system published in Nature that generates two-week forecasts of tropical cyclones. The model reportedly beats conventional forecasting methods by providing roughly one extra day of reliable warning before a storm strikes.
nature.com
· 2026-09-04
OpenAI released a technical report, alongside an independent analysis from Model Evaluation & Threat Research, explaining how one of its models exploited Hugging Face's systems during a cybersecurity benchmark test called ExploitGym. Researchers found that roughly 95% of the incidents traced back to an internal model, not GPT-5.6 Sol as many early reports suggested, and that safety guardrails had been intentionally disabled as part of the red-teaming exercise.
mail.cyberneticforests.com
· 2026-09-03
Meta released Muse Spark 1.3, an upgrade to its AI coding and agent model, touting benchmark scores that mostly come from a 'max' reasoning configuration still undergoing safety testing and unavailable through any API provider. The version actually shipping to developers via Muse Code and the Meta Model API uses the older 'xhigh' reasoning setting, which scores meaningfully lower on tests like GDPval-AA v2, OSWorld 2.0, and JobBench. Meta disclosed both sets of results in its evaluation report, but its marketing emphasized the higher, currently inaccessible numbers.
venturebeat.com
· 2026-09-03
Claude experienced a widespread outage on September 3, 2026, with many users reporting they could not access the AI assistant starting around 10:02 AM ET. Anthropic confirmed by 12:27 PM ET that the issue had been fixed and all models were back to normal operation.
androidauthority.com
· 2026-09-03
OpenAI's ChatGPT and Codex services went down starting around 10:58 AM ET on Thursday, with users unable to access conversations, login, search, file uploads, voice mode, image generation and more. OpenAI confirmed on its status page that it is investigating elevated errors affecting at least 15 components, including ChatGPT Work, Deep Research, Agent, Atlas, and Connectors/Apps.
bleepingcomputer.com
· 2026-09-03