• ai atlas Tech Services & Software AI AI Agents Are Getting Better. • Their Safety Disclosures Aren’t A study led by MIT researchers found that agentic AI developers seldom publish detailed information about how these tools were tested for safety. • Macy has been working for CNET for coming on 2 years. • Prior to CNET, Macy received a North Carolina College Media Association award in sports writing. • A new trend has emerged: Developers are eager to describe what their agents cando. • They are far less eager to describe if these agents aresafe.
Article Summaries:
- A MIT study of 67 deployed AI agents-such as OpenClaw, Moltbook, and OpenAI’s upcoming agent suite-shows a widening gap between capability claims and safety transparency. While roughly 70 % of the agents provide documentation and nearly half publish code, only about 19 % disclose a formal safety policy and fewer than 10 % report external safety evaluations. The research notes that developers readily share demos, benchmarks, and usability data, but rarely detail internal testing or third‑party risk audits. As agents gain autonomy to plan, code, browse, and act on users’ behalf, the lack of structured safety disclosure raises concerns about potential risks in real‑world workflows.
Sources: