AI Assistants Are Starting to Hack Daily Life
In Australia, a man named Andrew Bird tasked an AI assistant running Claude on Openclaw to book a spot in a consistently full pilates class. First, he noticed the assistant had booked sessions months out, far beyond what he expected. Then, when he asked whether it could move him up the waitlist, the assistant cancelled someone else’s reservation and enrolled Andrew in the class he wanted.
We think this is just a preview of a future arriving faster than most realize. Over the past decade of intense digitalization, small businesses building their systems likely prioritized speed and cost over security, understandably so, since the number of people with the skills to actually exploit these gaps was small and unlikely to bother. So this gym put its security layer only on the frontend: discover the server address and you can make any change you want, no authorization required in the backend. Finding these kinds of vulnerabilities is child’s play for today’s language models, and as these assistants enter our daily lives, the gaps in poorly designed systems will surface one by one.
OpenAI’s Cybersecurity-Specific Model
Right on cue, OpenAI announced GPT-5.6-Cyber, a model customized for cybersecurity, saying the future of AI-driven cyberattacks is approaching fast and defense mechanisms need to be strengthened. In real-world tests, the model found two previously unknown vulnerabilities in Chrome’s V8 engine. Access requires identity verification and legal commitments.
Meta’s New Open-Source Model: Muse Glimmer
Meta announced Muse Glimmer, a new 30-billion-parameter model built specifically for on-device AI automation. A compact version of Meta’s most advanced model, Spark, Muse Glimmer runs on a single GPU with no internet connection required. Meta’s willingness to compete against Chinese models, and its refusal, unlike American rivals, to abandon open source, is commendable. That said, Meta’s real goal is to develop models that support its own hardware so it can continue making money from advertising and devices.
This Week’s Inspiration
If you feel uncertain about AI’s impact on creativity and art, this SXSW talk by Patreon CEO Jack Conte might help you find your footing. The presentation design itself is striking and original; we enjoyed it thoroughly.
⚡️ Quick Bits
• A team that cracked the reasoning chains of OpenAI, Anthropic, and Google’s language models using weaker models says the exposed data includes users’ API keys and personal information. Stolen Thoughts →
• Google DeepMind is reshuffling. Chief Scientist Jeff Dean is leaving after 27 years to start his own company, and CEO Demis Hassabis is stepping down. CNBC →
• GrokBot has joined the autonomous AI assistants. X →
• Claude will now add watermarks to all the text it generates. Claude →
📩 Liked this newsletter? You can forward it to your friends
if you are feeling nostalgic, though it’s always better if
they just subscribe on their own.
