OpenAI launched ChatGPT Atlas, the latest AI-powered web browser with “agent mode” that can autonomously navigate websites and complete multi-step tasks on users’ behalf.

Via Artificial Ignorance

https://www.ignorance.ai/p/ai-roundup-the-browser-wars-continue

“The browser wars continue

OpenAI launched ChatGPT Atlas, the latest AI-powered web browser with “agent mode” that can autonomously navigate websites and complete multi-step tasks on users’ behalf.

The big picture:

• Atlas builds on OpenAI’s previous agentic AI attempts like Operator and ChatGPT Agent, featuring personalized memory, a persistent ChatGPT companion in split-screen mode, and “cursor chat” for in-line text editing. But early testing reveals a slow, clunky experience, with significant limitations on session length.
• The security implications are substantial, as even OpenAI’s CISO acknowledged: prompt injection attacks remain an “unsolved frontier security problem” that could allow malicious websites to hijack the agent and leak credentials.
• By launching a browser, OpenAI is directly competing with Google, Microsoft, and Apple on their home turf – not to mention Perplexity’s Comet and Opera’s Neon. However, whether users want an AI intermediary for everyday web tasks – and whether they’ll trust it with sensitive actions – remains an open question.”