Weekly News #31 | Updates Worth Your Attention 👀
Anthropic announced that they have detected 16 million suspicious transactions by 24000 fraudulent accounts on Claude. The company claims these accounts were linked to Chinese competitors DeepSeek, Moonshot and MiniMax. According to Anthropic, the accounts were used to improve rival AI models and their access have now been blocked.
We all know that OpenAI has been quietly working with former Apple chief designer Jony Ive on smart devices for a while. Last week’s leak claimed that their first product could be a camera-equipped smart speaker.
This move could heat up competition in the smart home assistant market. Amazon has yet to fully integrate state-of-art AI into Alexa and OpenAI now seems ready to introduce a direct competitor. If these rumors are true, OpenAI is getting ready for an aggressive entry into the hardware space.
One Vacuum to Rule Them All
Developer Sam Azdoufal just wanted to use his DJI robot vacuum with an Xbox controller. To access the device’s closed environment, he used Claude Code and the token he got unlocked his own vacuum’s API. Plus, 7000 others worldwide. 😶🌫️
Azdoufal reported this vulnerability to DJI and the company has since fixed the issue. Yet the incident raised serious concerns about the security of smart home devices and what could have happened if the discovery had been made by someone with bad intentions.
This Week in “Bot Incidents of Silicon Valley”
Summer Yue, director of AI safety and alignment at Meta Superintelligence, showed us a new way of what happens when you give an autonomous bot too much access. After allowing OpenClaw to manage her inbox, the bot started deleting her emails. Attempts to stop it remotely via Telegram did nothing. She ultimately had to run to the Mac mini running the bot and shut it off manually.
Sometimes the only kill switch is physical. 🗡️
FastCompany → | X/Summer Yue →
tio
2/25/2026
