OpenAI: Difference between revisions

Line 28:

===ChatGPT Atlas and prompt-injection vulnerability (2025)===

In 2025, Brave posted an article about vulnerabilities that have agentic web browsers, such as ChatGPT Atlas, that consists of adding hidden malicious prompts in files, text or another media. Those prompts, combined with weak safeguards of the AI agents, can make them to expose and leak sensitive data of the user.<ref>https://owasp.org/www-community/attacks/PromptInjection ([http://web.archive.org/web/20260210124436/https://owasp.org/www-community/attacks/PromptInjection Archived])</ref>

In 2025, Brave posted an article about vulnerabilities that have agentic web browsers, such as ChatGPT Atlas, that consists of adding hidden malicious prompts in files, text or another media. Those prompts, combined with weak safeguards of the AI agents, can make them to expose and leak sensitive data of the user.<ref>https://owasp.org/www-community/attacks/PromptInjection ([http://web.archive.org/web/20260210124436/https://owasp.org/www-community/attacks/PromptInjection Archived])</ref><ref>{{Cite web |last=Kaul Sahib |first=Shivan |last2=Chaikin |first2=Artem |date=21 Oct 2025 |title=Unseeable prompt injections in screenshots: more vulnerabilities in Comet and other AI browsers |url=https://brave.com/blog/unseeable-prompt-injections/ |url-status=live |archive-url=https://web.archive.org/web/20251027012608/https://brave.com/blog/unseeable-prompt-injections/ |archive-date=2025-10-27 |access-date=14 Jun 2026 |website=Brave}}</ref>

===Funding of the Parents & Kids Safe AI Act and creation of a child safety organization (2026)===

Line 51:

===Misc.:===

*[[ChatGPT]]

*ChatGPT Atlas

*ChatGPT Atlas browser

==See also==

@@ Line 28: / Line 28: @@
 ===ChatGPT Atlas and prompt-injection vulnerability (2025)===
-In 2025, Brave posted an article about vulnerabilities that have agentic web browsers, such as ChatGPT Atlas, that consists of adding hidden malicious prompts in files, text or another media. Those prompts, combined with weak safeguards of the AI agents, can make them to expose and leak sensitive data of the user.<ref>https://owasp.org/www-community/attacks/PromptInjection ([http://web.archive.org/web/20260210124436/https://owasp.org/www-community/attacks/PromptInjection Archived])</ref>
+In 2025, Brave posted an article about vulnerabilities that have agentic web browsers, such as ChatGPT Atlas, that consists of adding hidden malicious prompts in files, text or another media. Those prompts, combined with weak safeguards of the AI agents, can make them to expose and leak sensitive data of the user.<ref>https://owasp.org/www-community/attacks/PromptInjection ([http://web.archive.org/web/20260210124436/https://owasp.org/www-community/attacks/PromptInjection Archived])</ref><ref>{{Cite web |last=Kaul Sahib |first=Shivan |last2=Chaikin |first2=Artem |date=21 Oct 2025 |title=Unseeable prompt injections in screenshots: more vulnerabilities in Comet and other AI browsers |url=https://brave.com/blog/unseeable-prompt-injections/ |url-status=live |archive-url=https://web.archive.org/web/20251027012608/https://brave.com/blog/unseeable-prompt-injections/ |archive-date=2025-10-27 |access-date=14 Jun 2026 |website=Brave}}</ref>
 ===Funding of the Parents & Kids Safe AI Act and creation of a child safety organization (2026)===
@@ Line 51: / Line 51: @@
 ===Misc.:<!-- Placeholder. -->===
 *[[ChatGPT]]
-*ChatGPT Atlas
+*ChatGPT Atlas browser
 ==See also==