Cybersecurity researchers from Hacktron successfully accessed OpenAI user accounts by leveraging Anthropic's Claude models to identify and exploit software vulnerabilities.
Key facts
- •Hacktron researchers used Anthropic's Claude models to exploit a vulnerability in the Discourse service used by OpenAI's community forum.
- •The breach allowed unauthorized access to ChatGPT and Codex accounts, including those belonging to OpenAI staff.
- •OpenAI launched GPT-6 Sol and Luna, which are 50% cheaper than the previous GPT-5.6 promotional pricing.
- •Anthropic's new Claude Opus 5.5 model is reported to be 40% more cost-efficient than the Opus 5 version.
- •Industry leaders, including Dario Amodei and Sam Altman, have recently called for a slowdown in advanced AI development.
Three cybersecurity researchers from the firm Hacktron gained unauthorized access to OpenAI community forum accounts by using Anthropic's Claude AI models to identify security flaws. The team, led by Mohan Pedhapati, utilized the Claude Opus 4.8 model to locate vulnerabilities in the Discourse service used by the forum, eventually succeeding with the more advanced Claude Opus 5. This breach allowed the researchers to access ChatGPT and Codex accounts, including those belonging to OpenAI employees.
By the numbers
The Security Breach Details
In late July, the Hacktron team targeted the OpenAI community forum, where users sign in with their OpenAI credentials. After initial attempts to exploit a flaw in the third-party Discourse service using Claude Opus 4.8 failed, the team successfully gained access the following day after the release of Claude Opus 5. Once inside, the researchers could view private ChatGPT conversations and potentially access connected applications, such as email services and Slack, which were linked to the compromised accounts.
New Model Releases and Market Competition
Following the incident, both OpenAI and Anthropic announced new, more cost-effective AI models. OpenAI introduced GPT-6 Sol and GPT-6 Luna, which are priced at 50% less than the GPT-5.6 family. These models are designed for specific workloads, with Sol targeting complex tasks like coding and Luna focused on high-volume information extraction. Simultaneously, Anthropic released Claude Opus 5.5, which the company claims is 40% more cost-efficient than its predecessor due to reduced token usage.
Industry Safety and Development Debates
The recent product rollouts occur amid an ongoing industry debate regarding the pace of AI development and safety protocols. Anthropic CEO Dario Amodei, along with OpenAI CEO Sam Altman and Elon Musk, have recently called for a slowdown in the development of advanced AI. This discourse was prompted in part by former Anthropic researcher Jacob Coxon, who resigned on September 8, warning that major AI labs are taking significant risks with current development trajectories.
Timeline
- Late JulyHacktron researchers used Claude models to identify vulnerabilities in the OpenAI community forum.
- September 8Former Anthropic researcher Jacob Coxon announced his resignation and warned about AI safety risks.
- TuesdayOpenAI and Anthropic both announced new, lower-cost AI model tiers.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by Ars Technica, CBS News Technology, CNBC Technology.
