OpenAI has introduced Images 2.6 to ChatGPT, Codex, and the API, enabling faster image generation and enhanced control over editing. This new version allows for the creation of GIF animations, bringing motion to ideas such as animating brand mascots or transforming product shots into eye-catching loops.

Images 2.6 generates images 50% faster than its predecessor, offering two models in the API: GPT-Image-2.6 Flare and Sunburst, depending on quality settings. Flare can generate images in as little as six seconds at low quality, while Sunburst provides higher quality and greater consistency across edits.

Developers gain more precise control over image creation and editing through features like Image markup and SVG exports. A new complexity parameter also allows adjusting the model's planning effort for a request, with lower settings suitable for quick edits.

👉 Read the full article on OpenAI

OpenAI has unveiled an autonomous pipeline for discovering, exploiting, and patching vulnerabilities in the Linux kernel. In a single run, this system identified 2,308 potential vulnerabilities, proved 948 of them to be genuinely dangerous, and generated the necessary patches.

The API token cost for one solitary run of this massive AI workflow, if billed to a client, would amount to $1,213,430, according to OpenAI. However, the company calculates this as a bargain, representing a 16x to 55x return on investment compared to the $18.9 million to $66.3 million typically paid to white-hat hackers in bug bounties.

This effort mirrors a smaller-scale initiative previously undertaken as part of their "Patch the Planet" program. The trend of frontier AI labs focusing on vulnerability discovery is also highlighted by Google's claim last year of using an automated tool called Big Sleep to stop an attack in progress.

👉 Read the full article on Gizmodo

OpenAI has developed an autonomous pipeline for discovering, exploiting, and patching local privilege escalation (LPE) vulnerabilities in the Linux kernel. In a single run, this pipeline validated and generated patches for 948 LPE vulnerabilities, incurring an API cost of $1.2 million.

This research demonstrates that vulnerability discovery can increasingly be run as a systematic process, highlighting the effectiveness of frontier models in identifying security flaws and demonstrating their impact in large, critical codebases like the Linux kernel. The pipeline consists of five stages: vulnerability discovery, dynamic validation, exploitability assessment, patch generation, and human review.

OpenAI's analysis suggests that the 948 LPE exploits found by this pipeline have a hypothetical valuation of $18.9 million to $66.3 million when compared to historical bug bounties, approximately 16x to 55x the pipeline's cost. Furthermore, the AI agents reportedly discovered at least 10 novel Linux exploitation techniques.

The research suggests the importance of rescanning previously audited code as model capabilities improve, as stronger models can uncover vulnerabilities missed by earlier versions. It also indicates that more capable models require lighter orchestration and that the quality of AI-generated patches improves with proper instruction and iteration.

👉 Read the full article on OpenAI