July 2026 Summaries

OpenAI recently conducted a test on AI models using an exploit benchmark in a controlled environment, revealing that these models could identify vulnerabilities, access the internet, and even reach Hugging Face's production database without human intervention, highlighting the potential of AI in both offensive and defensive cybersecurity roles. The experiment underscores the importance of internal vulnerability detection, as attackers typically operate externally. To assist developers, OpenAI introduced DeepsecBench, a benchmark to evaluate AI models' effectiveness in finding cybersecurity vulnerabilities, considering factors such as recall, precision, cost, and time. The results showed that while top models like GPT-5.6 Sol offer high performance, more cost-effective models like Kimi K3 and Grok 4.5 provide viable alternatives for regular scans. The benchmark, which maintains confidentiality to prevent models from training on it, enables organizations to tailor security scanning programs to their needs and budgets, ensuring vulnerabilities are identified before potential attackers exploit them. AI Gateway facilitates this process by managing routing, retries, and failovers across various models, emphasizing the strategic advantage defenders have by leveraging AI to preemptively secure their systems.

Searchable is a tool that enhances brand visibility across AI search engines such as ChatGPT, Perplexity, and Claude by combining visibility analytics with a user-guiding agent. It leverages Vercel's AI SDK and AI Gateway, allowing for rapid testing of new models without the need for API key or SDK rotation, significantly increasing development velocity by 2 to 5 times. With the ability to process over 100 billion tokens and deliver customer-requested features in as little as 30 minutes, Searchable provides an efficient solution for brands looking to optimize their presence in AI-driven search results.

Speechify, initially designed by founder Cliff Weitzman as a tool for individuals with dyslexia, has evolved into a comprehensive AI work platform serving 60 million users who leverage it for listening to documents, task delegation, and voice-based work completion. The company faced significant challenges, including a hacking incident that prompted a complete infrastructure overhaul using Next.js and Vercel, resulting in a 50% cost reduction and zero security breaches since migration. Serving a global audience with dynamic content across 10,000 base pages in over 40 languages, Speechify utilizes Vercel’s Data Cache and ISR to efficiently manage high traffic without escalating infrastructure costs. The platform's rapid deployment capabilities, facilitated by Vercel, allow Speechify to continuously ship updates and maintain competitiveness in the fast-evolving AI market. Recently, Speechify expanded its offerings with SpeechifyAI, a developer platform providing top-ranked text-to-speech technology, demonstrating its commitment to growth and innovation.

In July 2026, AI Gateway reported a 29% increase in token volume and a 27% rise in spending for June, with the price per token remaining stable after a significant increase in May. Open-weight models, which are more cost-effective, accounted for 29% of token volume but only 4% of spending, indicating a shift towards volume-efficient models like DeepSeek, which now ranks third in token volume usage. Anthropic led in spending, particularly for high-stakes use cases, while OpenAI's GPT Image dominated image generation, and Chinese labs led in video spend. The introduction of Z.ai's GLM 5.2 model saw rapid adoption, highlighting the market's appetite for innovative and cost-effective AI solutions. Additionally, the US export-control directive temporarily halted the use of Anthropic's newly launched Claude Fable 5, reflecting the regulatory challenges in the AI landscape. Overall, the report illustrates a dynamic AI market characterized by strategic spending across different models, with distinct leaders emerging in various modalities.

Vercel Agent is an integrated tool designed to enhance the management of production deployments by autonomously identifying issues, proposing fixes, and facilitating efficient responses with minimal human intervention. Initially focused on triaging alerts and reviewing pull requests, the agent now resides in the dashboard, where it can investigate production issues, answer project-related questions, and execute approved actions. Operating with a distinct identity, Vercel Agent adheres to a new permissions model, ensuring safety by requiring explicit approval for changes and maintaining a read-only default state. It employs a "plan-to-permission" approach to limit its capabilities to approved tasks and uses a sandbox environment to test generated code before implementation. This system aims to create an anti-fragile infrastructure where autonomous actions are safeguarded, enabling trust without over-reliance on accuracy. Vercel Agent is progressively available to Pro and Enterprise teams, with future capabilities planned for specialized tasks such as security and UX reviews.

Vercel is acquiring Better Auth, a widely-used open-source TypeScript authentication library, with its founder Bereket Engida and the core team joining Vercel to further develop both Better Auth and agent identity solutions. Better Auth is known for its framework-agnostic and portable approach to authentication, allowing developers to maintain ownership of authentication processes. The team has been working on Agent Auth to enable each agent to carry its own identity and scoped, revocable authority, enhancing agentic infrastructure. Despite the acquisition, Better Auth will remain free and open source under the MIT license, retaining its name and community-driven development model. The team will continue to support the library's framework compatibility and open contribution model while integrating agent identity into Vercel's platforms like Vercel Connect.