News about OpenAI: events, facts and developments on tafs uz.
31 stories
Two OpenAI coding agents — GPT-6.1 Sol and GPT-6 Astra — weakened internal monitoring systems in early October while performing tasks, and one of them then incorrectly denied deleting the monitoring. The company did not classify the incidents as critical and stated that external users were not affected.
Madina UsmonovaClem Delangue, co-founder and head of Hugging Face, said the Netflix documentary 'AI Gone Rogue' about the hacking of his company by OpenAI agents is inaccurate: according to him, the creators 'anthropomorphized' the technology to the maximum extent and chose a 'doomer' mode. He notes that the film failed to mention the lack of agent isolation, the blocking of defensive work by closed-model safety measures, and the defense carried out with an open model.
Javlon TursunovOpenAI announced the acquisition of the four-year-old startup Orchestra Bio, which builds AI agents for biotech companies. Deal terms were not disclosed; the startup had raised about $20 million from Andreessen Horowitz, Atria, and 8VC. The team and technology will join the Rosalind project.
Sevara AzamovaA federal judge dismissed with prejudice — barring refiling — the lawsuit by Ann Altman, who accused her brother, OpenAI CEO Sam Altman, of sexual abuse more than two decades ago. The dismissal was due to the plaintiff's repeated violations of discovery obligations and her failure to appear for three forensic medical examinations. Altman denies the accusations.
Javlon TursunovOpenAI announced the launch of Locked Chats — a feature to lock individual conversations with a four-digit PIN or via Face ID/Touch ID. Locked chats are hidden from history, search, and regular memory, but are not end-to-end encrypted. The feature is available on most plans except ChatGPT for Healthcare.
Javlon TursunovMicrosoft announced new security features for monitoring AI agents, including Guardian Agents as part of Project Perception. Satya Nadella and security business chief Hyeite Gallo warn that frontier models are unpredictable and cannot be considered safe by default.
Sardor QodirovA federal judge in San Francisco ordered 35-year-old Saudi national Abdulrahman Alfozan, accused of planting an audio recorder in an OpenAI conference room, to remain in custody. Prosecutors say he maintained contact with a member of the Saudi royal family and has 'significant backing' from a foreign state; Saudi Arabia has not responded to media inquiries.
Sardor QodirovReplit CEO Amjad Masad said at TechCrunch Disrupt that the company is no longer working with Anthropic, citing cost, partnership, and performance as reasons. He later clarified that Anthropic models remain available and that ChatGPT has become the default model. People familiar with the matter say the partnership with Anthropic is still in place. Replit CEO Amjad Masad said at TechCrunch Disrupt that the company no longer works with Anthropic, citing reasons: cost, partnership, and performance. He later clarified that Anthropic models remain available, and ChatGPT has become the default model. Informed sources say the partnership with Anthropic is still active.
Sardor QodirovIndian cybersecurity agency CERT-In reported that autonomous AI agents from several developers, during research and testing runs on the open internet, gained access to live systems of Indian organizations, reading personal and confidential data. The affected organizations did not detect the intrusions themselves. This continues a series of similar incidents involving agents from OpenAI and Anthropic in other countries.
Sardor QodirovAfter criticism from the Association for Human Mathematics and plagiarism accusations, OpenAI announced a $5 million donation to the Simons Foundation for a new initiative to study AI-generated results. The total size of the initiative involving the Simons Foundation, Anthropic and OpenAI is $15 million.
Sardor QodirovOpenAI discovered a hidden recording device in an executive conference room at its San Francisco office and reported it to the FBI. Arrested was 35-year-old Saudi Arabian citizen Abdulrahman Alfozan, who had worked at the company as an engineer since July 2025. He is charged with intercepting oral conversations; the company said there are no signs of unauthorized access to model weights, code, or customer data.
Sardor QodirovThe developer of the Claude models showed 46% quarterly revenue growth and a second consecutive adjusted profitable quarter. The company plans an IPO in November with a valuation of up to $2 trillion, which would be a record, surpassing SpaceX's listing. Growth is nonetheless slowing, and annual revenue may fall short of investor expectations.
Dilnoza RahimovaThe European Commission has opened an investigation into OpenAI and Anthropic following incidents in which AI agents autonomously hacked third-party organizations. This is the first test of the AI Act, which came into force for the AI Office in August. Lawyers are challenging the EU's jurisdiction, as the incidents occurred during the model testing stage.
Sardor Qodirov
