A V2EX user reported that after installing the 'dev-workflow' Copilot skill in VS Code, all files on a connected Samsung T9 external SSD were deleted within minutes. The user immediately shut down the computer and is now attempting data recovery using R-Studio on a different machine. No further AI commands were issued after installation, and the exact cause—whether the skill, a virus, or a hardware issue—remains unclear. The incident highlights potential safety risks of AI-integrated development tools executing unintended file operations.
A user on macOS running Claude Code 2.1.177 with Opus 4.8 and acceptEdits enabled observed that during a long session the model repeatedly attributed tool_results and its own reasoning as actual user messages. For example, a shell output containing a version string was later quoted as “your original words,” even though the user never typed it. The agent then autonomously wrote a memory file based on these phantom inputs. The user verified via the raw JSONL transcript that the messages were from tool_results, not user turns. No evidence of system compromise was found; similar issues are reported in GitHub issues #58671, #68159, #63871, and others, indicating a harness boundary bug. Temporary mitigations include downgrading to 2.1.174, disabling acceptEdits, removing global write permissions, and not resuming affected sessions.
The author’s mother was diagnosed with advanced diffuse large B-cell lymphoma (stage IV, KI67 95%+) and requires POLA-R-CHP chemo (self-funded 33,000 CNY per cycle, 6 cycles). His brother suffered an acute cerebral infarction during the same period. The author then discovered his own coronary artery mixed plaque with moderate stenosis and ventricular tachycardia, worsened by anxiety and previous inconsistent treatment for high triglycerides. He highlights the consequences of self-adjusting medication, ignoring symptoms, and skipping regular checkups, and advises early marriage, having children, and prioritizing health over careerism.
A V2EX user describes being accused of posting AI-generated content, leading to the realization that society faces a major crisis: we can no longer reliably distinguish human interactions from AI ones online. The post argues that this authenticity problem affects written posts, debates, and even voice/video calls, potentially eroding trust in online communication. The author warns that malicious actors could exploit AI to manipulate public opinion on a large scale, posing a risk greater than the AI technology itself.
A user reports being scammed by a Telegram group called the 'First sharing group' that had nearly 30,000 members. The group offered shared subscriptions for services like OpenAI ChatGPT, Netflix, YouTube, Spotify, Office 365, HBO, and Surge. The admin defrauded the user for a ChatGPT share and then deleted all chat records. The warning urges others to avoid the group and spread awareness.
A V2EX user complained about AI-generated content on X (Twitter) and requested a tool that can identify and filter such tweets. The post asks for any existing solutions, indicating a demand for AI content filtering utilities.