• Is a secure AI assistant possible? • Experts have made progress in LLM security. • But some doubt AI assistants are ready for prime time. • AI agents are a risky business. • Even when stuck inside the chatbox window, LLMs will make mistakes and behave badly. • Once they have tools that they can use to interact with the outside world, such as web browsers and email addresses, the consequences of those mistakes become far more serious.
Article Summaries:
- Experts say progress in large‑language‑model (LLM) security is underway, but doubts remain about the readiness of AI assistants for everyday use. In November 2025 independent engineer Peter Steinberger released OpenClaw, a GitHub‑hosted tool that lets users build custom assistants with any LLM. The project quickly went viral, but it also raised serious privacy and security concerns: users must grant the assistant access to emails, files, and payment data. Security researchers have identified multiple vulnerabilities, prompting a public warning from the Chinese government. Steinberger cautioned non‑technical users against using the software, while the industry faces a clear need to develop robust, data‑safe personal‑assistant platforms.
Sources: