Anthropic is escalating the battle to control how people use the internet, rolling out browser technology that could make Google Chrome less essential for

https://exa.ai/blog/exa-firefox

AI has changed how we browse. Full stop. The real question is whether AI leads to one browser experience for everyone, or gives people more ways to make th

Atlas just shut down. Compare the 3 real replacements — ChatGPT's Chrome extension, Perplexity Comet, and Sigma Browser.

A former Perplexity employee who worked on the Comet browser has spun out to build Polar, an AI-native browser targeting knowledge workers. The startup has raised a $5.7 million seed round led by Madrona.

Polar has raised $5.7 million to build an #AIbrowser that automates workplace tasks, reflecting a broader industry shift away from AI-powered search and toward browser agents.

올해 들어 IT 업계에서 가장 뜨거운 화제는 단연 인공지능입니다. 특히 단순히 검색 결과를 보여주던 시대를 넘어, 사용자 대신 직접 웹사이트를 탐색하고 할 일을 수행하는 AI 웹 브라우저들이 속속 등장하고 있습니다. 구글 크롬이나 애플 사파리를 쓰고 계신다면, 이제 브라우저가 단순한

AI News: OpenAI has officially decided to retire Atlas, the AI-powered browser it introduced in October with ChatGPT at its center, marking a significant

OpenAI has announced the discontinuation of its AI browser "Atlas", transferring its agent features to the ChatGPT desktop app and Chrome extension, marking a strategic pivot away from the browser war

OpenAI has announced that Atlas, its standalone AI-powered desktop browser, will be discontinued. In its place, the company is building a unified ChatGPT desktop application. Codex has already been integrated into ChatGPT, and Atlas-related browser features will follow in the same direction. Integration in Name, Retirement in Practice This integration story is more nuanced than it first appears. The latest version of ChatGPT, now centered on Codex, has shifted the app's focus considerably.

OpenAI will retire the ChatGPT Atlas browser on August 9, moving its AI browsing features into the redesigned ChatGPT desktop app.

AI browsers are taking on increasingly ambitious roles as autonomous agents. Yet the more they act on a user's behalf, the more dangerous blind trust in page context becomes. Researchers at LayerX have described a new attack technique called BioShocking. It tricks an AI browser into accepting fabricated rules from a malicious page. The technique then pushes the agent beyond its own protective guardrails.

AI browsers are meant to make you safer online. The BioShocking attack uses that intelligence against you — and most users have no idea it's happening.

 Over ten million people might face major online threats following the discovery of severe weaknesses in two common AI-based Chrome add-ons, SiderAI and MaxAI. Though designed to assist with summaries and automated tasks, these tools were found carrying dangerous bugs - dubbed “Spyder” and “MaXSS” - by analysts at Rebora Security during a routine check of such software. Once exploited, either flaw lets unauthorized parties hijack active browsing activities.  Information saved on sites, along with files on personal devices, may become reachable without permission. While built for convenience through side panels and smart responses, their broad adoption across Chromium-linked browsers amplifies how far harm could spread. Despite appearing helpful, the underlying structure allows invasive access when misused. One of the leading tools on the Chrome Web Store, SiderAI sits in the top quarter of all extensions by popularity.  A recent analysis revealed flaws in how SiderAI and MaxAI managed data flow between sites and their inner workings, especially involving content scripts. Although these scripts should serve as controlled messengers - keeping site code apart from backend logic - the boundaries blurred in practice. Messages sent by web pages entered without sufficient checks. Because verification steps were missing, untrusted inputs could move deeper into the system than intended. A flaw in MaxAI allowed harmful sites to transmit manipulated data directly to its content script.  Though meant to relay information, the system passed these signals onward - into the background process - with little checking. Because of this gap, unauthorized users gained access to powerful functions. Hidden tabs appeared without warning, snapshots of screens were captured, site interactions occurred - all while riding on logged-in accounts. Security weakened when trust was misplaced across internal components. Testing revealed researchers gaining entry to live Gmail and Google Calendar sessions, pulling confidential data while leaving no trace.  What made the Spyder vulnerability in SiderAI alarming was its ability to mimic real user behavior - clicks, typing - all within integrated browser windows. A compromised site, using this loophole, might load Google Gemini unseen, harvest ongoing AI dialogues, then send them outward. Detection during such an event remained unlikely. What happens because of these flaws goes well past messages or chat tools.  Through them, hackers might grab login codes, see private correspondence, change files, while acting like the victim on many sites. Sometimes, the broad access given to such add-ons lets intruders reach data saved directly on a person's device. What stands out most is how little effort an attacker needs - just opening a harmful webpage can trigger the flaw. Because of this low barrier, threats can spread fast without clear signs.  After uncovering the problem, Rebora Security reached out to the creators of the affected tools; silence followed. With no reply, the details eventually appeared online, while a heads-up also went to Google. Should SiderAI or MaxAI appear in a user's browser, removal is urgent. This case brings attention to rising risks tied to artificial intelligence add-ons - especially those collecting sensitive online behavior.  When apps gain deep access to personal information, careful review of their privileges becomes unavoidable. Security grows more complex as these tools spread across everyday browsing routines.

Discover the 5 essential things to know before switching to an AI browser like OpenAI ChatGPT Atlas or Perplexity Comet. Learn privacy tips, data risks, agentic features, comparisons, and safe usage i...