Claude से ChatGPT को हैक करने वाले इंडियंस कौन हैं? Hacktron OpenAI Hack

Three Indians Hack OpenAI's System

Overview of the Incident

• Three Indian researchers utilized approximately ₹.5 lakh in AI tokens to hack into OpenAI's system within just 72 hours.

• They accessed OpenAI’s source code by first infiltrating an employee's account, allowing them extensive control without causing any damage.

• This hacking occurred amidst warnings from AI leaders about potential threats posed by AI agents, highlighting a critical security lapse.

Discovery of the Bug

• The process began on July 23 when researchers identified a bug not directly in OpenAI but on its community discussion forum hosted by a third-party service called Discourse.

• They investigated how images uploaded to the forum were processed and noticed different handling methods for various image formats, particularly HEIC/HEIF used in iPhones.

Exploiting Security Flaws

• Researchers discovered that Discourse was using an outdated version of ImageMagick that lacked crucial security fixes, leading to vulnerabilities.

• The specific vulnerability was identified as a heap buffer overflow or out-of-bounds memory access bug, which allowed unauthorized data access.

Execution of the Attack

• On July 23, they crafted an image containing a message that triggered unlimited data access during decoding by the AI program.

• By July 24, after updating their tools to OPS 5, they successfully executed remote code on OpenAI’s community forum server.

Accessing User Accounts

• The researchers exploited another flaw where users could sign in via their OpenAI accounts through Discourse, linking both systems insecurely.

• They gained access to authentication tokens due to insufficient restrictions on user information retrieval from OpenAI’s servers.

Consequences and Reporting

• After accessing several ChatGPT accounts—including those belonging to OpenAI employees—they created a harmless pull request in the internal Git repository as proof of their breach.

• Their intent was not malicious; instead, they aimed to alert OpenAI about significant security vulnerabilities present in their systems.

Response from OpenAI

Findings and Rewards

• On July 25, OpenAI acknowledged two major issues: one related to Discourse and another within their own systems; both were subsequently fixed.

• The hackers received recognition and a reward of ₹5.5 lakh for identifying these bugs as part of responsible disclosure practices.

Background of Researchers

• The three individuals involved—Mohan Pehdapathi (co-founder), Harsh Jaiswal (co-founder), and Rahul Mani (researcher)—have extensive backgrounds in cybersecurity and bug hunting.

Implications for Cybersecurity

• This incident raised concerns regarding global cybersecurity standards compared to those implemented by countries like China, emphasizing vulnerabilities even among reputable organizations.