OpenAI年化收入达到426亿美元，两个月增长约29%
Related
OpenAI Releases Deployment Simulation Security Assessment Framework: Replaying Real Traffic Prediction to Reveal Alignment Risks Between GPT-5 Series and Intelligent Agents
According to Beating's monitoring, OpenAI has released a Deployment Simulation security assessment method to predict the risk of models going out of control in real-world environments before official release. The study replayed 1.3 million anonymized historical dialogues from the GPT-5 thinking model series between August 2025 and March 2026, and the median error in predicting the rate of unexpected malicious behavior was only 1.5 times. Traditional static evaluation sets (such as SWE-Bench) easily make the model aware of being in a test state, thus metagaming during testing. Experiments show that GPT-5.2's evaluation awareness rate is close to 100% under traditional benchmarks, but drops to 5.1% in deployment simulation traffic (5.4% in production environment), indicating that the model cannot distinguish the test state. The deployment simulation also successfully captured GPT-5.1's violation of using browser tools as a calculator before release (calculator hacking). For agent-based scenarios, OpenAI used 120,000 GPT-5.4 employee programming trajectories to simulate GPT-5.5 coded agent deployment. To avoid the security risks of directly manipulating real systems, the team introduced a tool simulator based on a large model, combined with read-only connections and a historical call database, improving the environment realism from 11.6% to 49.5%, making it virtually indistinguishable from real systems. Even for external auditors lacking private production traffic, the prediction error using the WildChat open-source dialogue dataset can be controlled within three times.
OpenAI has entered into a multi-year partnership with image copyright giant Getty, and will provide licensed images to ChatGPT.
Odaily Odaily reports that OpenAI has announced a multi-year image display partnership with image copyright giant Getty Images, integrating its licensed visual content into ChatGPT's search and discovery features. Getty Images previously sued an AI company over copyright issues related to AI training data, but the litigation has not progressed as expected. As AI-generated content capabilities improve, the core value of media and copyright agencies is shifting from individual works to "compliant licensed content libraries" and distribution channels for content integrated into AI products. Copyright disputes will continue, but business cooperation models are becoming an important direction for the integration of the AI industry and the content industry. (Forbes)
OpenAI releases the full version of GPT-5.5-Cyber, achieving an 85.6% benchmark score, surpassing Mythos 5.
According to Beating's monitoring, OpenAI announced an upgrade to its Daybreak cybersecurity toolchain and officially launched the full version of the GPT-5.5-Cyber model for defenders. In the CyberGym benchmark test, which evaluates the ability of agents to reproduce known vulnerabilities, GPT-5.5-Cyber achieved the highest single-model score of 85.6%, surpassing GPT-5.5's 81.8% and competitor Mythos 5's 83.8%. With the widespread adoption of AI significantly improving vulnerability discovery efficiency, the bottleneck on the defense side has shifted from "finding vulnerabilities" to "automatic remediation." To translate detection results into actual remediation, the accompanying Codex Security plugin has been upgraded, allowing developers to automatically analyze and generate remediation patches within the Codex environment. Since its preview release in March, Codex Security has scanned over 30 million submissions and automatically confirmed the remediation of 500,000 security flaws. The latest version supports integration with CodeQL querying and exporting SARIF standard files, enabling the interception of potential vulnerabilities before software deployment. Meanwhile, OpenAI, in collaboration with Trail of Bits and HackerOne, launched the Patch the Planet open-source security project, providing ChatGPT Pro subscriptions and API quotas for over 30 mainstream open-source projects such as cURL and Go. Patches are manually verified by an expert team to reduce the maintainer's workload. In the Daybreak Partner Program, security vendors such as Palo Alto Networks and Wiz have been among the first to join, dedicated to implementing advanced defense capabilities in commercial security products.
Samsung Electronics' large-scale deployment of ChatGPT and Codex marks one of OpenAI's largest enterprise deployments to date.
According to Mars Finance, Samsung Electronics announced the deployment of ChatGPT Enterprise and Codex to all its employees in South Korea and its global Device Experience (DX) division. This collaboration represents one of OpenAI's largest enterprise-level deployments to date. Samsung Electronics plans to widely apply these AI tools across its core business areas, including software development, product development, marketing, and manufacturing, covering both technical and non-technical roles, to enhance employee productivity and problem-solving capabilities. This deployment also marks a significant expansion of Samsung's partnership with OpenAI, moving beyond early AI infrastructure (advanced memory semiconductor supply) to enterprise-wide AI-powered office applications.
OpenAI plans to launch ChatGPT advertising service in Japan.
According to Odaily Odaily, OpenAI plans to launch advertising features for its conversational AI service ChatGPT in the Japanese market. The ads will be targeted based on users' interests during chats. The advertising service is expected to be handled by agencies affiliated with Dentsu and Hakuhodo, marking a further expansion of the commercialization path for generative AI platforms. Sources indicate that the new ad format will be embedded in conversational interaction scenarios, matching ad content with users' real-time conversation topics and interests, thereby achieving more accurate content recommendations and commercial conversions. (Nikkei)
OpenAI updates GPT-5.5 Instant: Adaptive tone added, first rolled out to paid users.
According to Beating, OpenAI announced an upgrade to GPT-5.5 Instant, the most commonly used default model in ChatGPT. The new version focuses on enhancing the engaging nature of conversational interactions and improving the model's understanding of user intent and its ability to adapt to changing tone. The new model can more accurately identify the underlying intentions behind user questions and adjust its response style based on context, such as providing emotional support, practical advice, or in-depth analysis, to avoid providing mechanical, formulaic answers. In multi-turn conversations, the upgraded model can more reliably handle complex constraints and provide more practical and coherent product and local merchant recommendations. The update is initially rolled out to paid users (Pro, Plus, Business, Enterprise, and Go), with free users receiving the update gradually the following day.