Machine Learning Infrastructure Engineer

Nvidia Launches open-source AI Security alliance. Anthropic, OpenAI and Google Absent as 37 Firms Back Open AI Security Tools Chris Riotta (@chrisriotta) - July 27, 2026 Dozens of the world's biggest technology firms are joining a new Nvidia-led initiative to build open-source security tools for artificial intelligence. The U.S.-based chipmaker announced the "new alliance" in a Monday blog post, saying companies like IBM, Palantir and Microsoft "will work to remediate and disclose vulnerabilities using open technologies" as part of an effort to better safeguard software and AI agents. The Open Secure AI Alliance is a coalition of 37 firms, including CrowdStrike, Palo Alto Networks, Dell Technologies, Hugging Face and the Linux Foundation. "Open models, like any powerful technology, can be misused - including through attempts to weaken safeguards or repurpose capabilities for cyberattacks," the blog post read. Nvidia added that those risks "must be managed wherever advanced AI is deployed." The launch comes as Washington is weighing restrictions on open-weight AI models, and just days after a breach at Hugging Face sparked industry debate over whether closed AI systems can slow incident response efforts. While the initiative includes many of the biggest names in cloud computing, enterprise software and cybersecurity, it lacks several key players in the global AI sector, including Anthropic, OpenAI and Google - developers behind some of the industry's most capable closed models. The Nvidia blog post directly referenced the security incident Hugging Face disclosed earlier this month, in which OpenAI models running on an internal hacking benchmark with manipulated cyber safety refusals escaped a test environment and gained the ability to run commands on the AI repository's production servers. Commercial closed models declined to assist forensic analysis of the intrusion, which Nvidia said left the tools unable to tell apart the attackers from defenders. Hugging Face instead turned to GLM 5.2, an open-weight model from Chinese developer Z.ai. Defenders who cannot inspect and run advanced AI on their own systems face significant constraints during cyber incidents, Nvidia said. The alliance is urging policymakers to treat open models, harnesses and security tooling as "defensive assets, not liabilities" and warns that blanket restrictions on open-frontier systems would continue to concentrate power and dependence on a handful of closed providers. The announcement also follows a separate industry letter released Friday in which Nvidia, Microsoft, Meta and dozens of other companies urged the White House to avoid premature restrictions on open-weight AI. The list of signatures nearly doubled to 50 within a day as OpenAI and Google added their names over the weekend, while Anthropic and Amazon did not participate in both efforts. Anthropic has not publicly explained its absence from the alliance or the letter, and it is unclear whether the company was invited to join either. Nvidia and the alliance's cloud, enterprise software and security firms benefit from cheaper, customizable open models, while frontier firms like Anthropic and OpenAI earn revenue by selling access to their closed systems. The open-versus-closed debate also intensified last week after White House officials accused Chinese AI firm Moonshot AI of stealing American intellectual property through model distillation. Treasury Secretary Scott Bessent threatened sanctions against Chinese developers, while the Trump administration is reportedly weighing a ban on Chinese open-source models (see: AI Firms Seek US Help Against China Model Distillation). Nvidia is contributing open models, model weights, data and agent harness research to the effort through its Nvidia Labs Object-Oriented Agent project on GitHub. Microsoft is also offering MDASH, a multimodel scanning system that orchestrates specialized AI agents to discover and prove exploitable bugs, while SpaceXAI has open sourced its Grok Build coding agent and plans to release the weights of its Grok model line. Monday's launch is the latest in a string of industry safety coalitions. Anthropic, Google, Microsoft and OpenAI founded the Frontier Model Forum in 2023 to coordinate on responsible frontier development, and the Biden administration enlisted more than 200 organizations in a federal AI safety consortium the following year (see: White House Launches First-Ever AI Safety Consortium). Nvidia noted in its blog that the group is not opposed to closed systems, saying the world needs closed and open models working together so defenders can choose the right tool for the job. Nvidia and Anthropic did not immediately respond to requests for comment.

Nvidia Launches open-source AI Security alliance. Anthropic, OpenAI and Google Absent as 37 Firms Back Open AI Security Tools Chris Riotta (@chrisriotta) - July 27, 2026 Dozens of the world's biggest technology firms are joining a new Nvidia-led initiative to build open-source security tools for artificial intelligence. The U.S.-based chipmaker announced the "new alliance" in a Monday blog post, saying companies like IBM, Palantir and Microsoft "will work to remediate and disclose vulnerabilities using open technologies" as part of an effort to better safeguard software and AI agents. The Open Secure AI Alliance is a coalition of 37 firms, including CrowdStrike, Palo Alto Networks, Dell Technologies, Hugging Face and the Linux Foundation. "Open models, like any powerful technology, can be misused - including through attempts to weaken safeguards or repurpose capabilities for cyberattacks," the blog post read. Nvidia added that those risks "must be managed wherever advanced AI is deployed." The launch comes as Washington is weighing restrictions on open-weight AI models, and just days after a breach at Hugging Face sparked industry debate over whether closed AI systems can slow incident response efforts. While the initiative includes many of the biggest names in cloud computing, enterprise software and cybersecurity, it lacks several key players in the global AI sector, including Anthropic, OpenAI and Google - developers behind some of the industry's most capable closed models. The Nvidia blog post directly referenced the security incident Hugging Face disclosed earlier this month, in which OpenAI models running on an internal hacking benchmark with manipulated cyber safety refusals escaped a test environment and gained the ability to run commands on the AI repository's production servers. Commercial closed models declined to assist forensic analysis of the intrusion, which Nvidia said left the tools unable to tell apart the attackers from defenders. Hugging Face instead turned to GLM 5.2, an open-weight model from Chinese developer Z.ai. Defenders who cannot inspect and run advanced AI on their own systems face significant constraints during cyber incidents, Nvidia said. The alliance is urging policymakers to treat open models, harnesses and security tooling as "defensive assets, not liabilities" and warns that blanket restrictions on open-frontier systems would continue to concentrate power and dependence on a handful of closed providers. The announcement also follows a separate industry letter released Friday in which Nvidia, Microsoft, Meta and dozens of other companies urged the White House to avoid premature restrictions on open-weight AI. The list of signatures nearly doubled to 50 within a day as OpenAI and Google added their names over the weekend, while Anthropic and Amazon did not participate in both efforts. Anthropic has not publicly explained its absence from the alliance or the letter, and it is unclear whether the company was invited to join either. Nvidia and the alliance's cloud, enterprise software and security firms benefit from cheaper, customizable open models, while frontier firms like Anthropic and OpenAI earn revenue by selling access to their closed systems. The open-versus-closed debate also intensified last week after White House officials accused Chinese AI firm Moonshot AI of stealing American intellectual property through model distillation. Treasury Secretary Scott Bessent threatened sanctions against Chinese developers, while the Trump administration is reportedly weighing a ban on Chinese open-source models (see: AI Firms Seek US Help Against China Model Distillation). Nvidia is contributing open models, model weights, data and agent harness research to the effort through its Nvidia Labs Object-Oriented Agent project on GitHub. Microsoft is also offering MDASH, a multimodel scanning system that orchestrates specialized AI agents to discover and prove exploitable bugs, while SpaceXAI has open sourced its Grok Build coding agent and plans to release the weights of its Grok model line. Monday's launch is the latest in a string of industry safety coalitions. Anthropic, Google, Microsoft and OpenAI founded the Frontier Model Forum in 2023 to coordinate on responsible frontier development, and the Biden administration enlisted more than 200 organizations in a federal AI safety consortium the following year (see: White House Launches First-Ever AI Safety Consortium). Nvidia noted in its blog that the group is not opposed to closed systems, saying the world needs closed and open models working together so defenders can choose the right tool for the job. Nvidia and Anthropic did not immediately respond to requests for comment.

Nvidia Launches open-source AI Security alliance. Anthropic, OpenAI and Google Absent as 37 Firms Back Open AI Security Tools Chris Riotta (@chrisriotta) - July 27, 2026 Dozens of the world's biggest technology firms are joining a new Nvidia-led initiative to build open-source security tools for artificial intelligence. The U.S.-based chipmaker announced the "new alliance" in a Monday blog post, saying companies like IBM, Palantir and Microsoft "will work to remediate and disclose vulnerabilities using open technologies" as part of an effort to better safeguard software and AI agents. The Open Secure AI Alliance is a coalition of 37 firms, including CrowdStrike, Palo Alto Networks, Dell Technologies, Hugging Face and the Linux Foundation. "Open models, like any powerful technology, can be misused - including through attempts to weaken safeguards or repurpose capabilities for cyberattacks," the blog post read. Nvidia added that those risks "must be managed wherever advanced AI is deployed." The launch comes as Washington is weighing restrictions on open-weight AI models, and just days after a breach at Hugging Face sparked industry debate over whether closed AI systems can slow incident response efforts. While the initiative includes many of the biggest names in cloud computing, enterprise software and cybersecurity, it lacks several key players in the global AI sector, including Anthropic, OpenAI and Google - developers behind some of the industry's most capable closed models. The Nvidia blog post directly referenced the security incident Hugging Face disclosed earlier this month, in which OpenAI models running on an internal hacking benchmark with manipulated cyber safety refusals escaped a test environment and gained the ability to run commands on the AI repository's production servers. Commercial closed models declined to assist forensic analysis of the intrusion, which Nvidia said left the tools unable to tell apart the attackers from defenders. Hugging Face instead turned to GLM 5.2, an open-weight model from Chinese developer Z.ai. Defenders who cannot inspect and run advanced AI on their own systems face significant constraints during cyber incidents, Nvidia said. The alliance is urging policymakers to treat open models, harnesses and security tooling as "defensive assets, not liabilities" and warns that blanket restrictions on open-frontier systems would continue to concentrate power and dependence on a handful of closed providers. The announcement also follows a separate industry letter released Friday in which Nvidia, Microsoft, Meta and dozens of other companies urged the White House to avoid premature restrictions on open-weight AI. The list of signatures nearly doubled to 50 within a day as OpenAI and Google added their names over the weekend, while Anthropic and Amazon did not participate in both efforts. Anthropic has not publicly explained its absence from the alliance or the letter, and it is unclear whether the company was invited to join either. Nvidia and the alliance's cloud, enterprise software and security firms benefit from cheaper, customizable open models, while frontier firms like Anthropic and OpenAI earn revenue by selling access to their closed systems. The open-versus-closed debate also intensified last week after White House officials accused Chinese AI firm Moonshot AI of stealing American intellectual property through model distillation. Treasury Secretary Scott Bessent threatened sanctions against Chinese developers, while the Trump administration is reportedly weighing a ban on Chinese open-source models (see: AI Firms Seek US Help Against China Model Distillation). Nvidia is contributing open models, model weights, data and agent harness research to the effort through its Nvidia Labs Object-Oriented Agent project on GitHub. Microsoft is also offering MDASH, a multimodel scanning system that orchestrates specialized AI agents to discover and prove exploitable bugs, while SpaceXAI has open sourced its Grok Build coding agent and plans to release the weights of its Grok model line. Monday's launch is the latest in a string of industry safety coalitions. Anthropic, Google, Microsoft and OpenAI founded the Frontier Model Forum in 2023 to coordinate on responsible frontier development, and the Biden administration enlisted more than 200 organizations in a federal AI safety consortium the following year (see: White House Launches First-Ever AI Safety Consortium). Nvidia noted in its blog that the group is not opposed to closed systems, saying the world needs closed and open models working together so defenders can choose the right tool for the job. Nvidia and Anthropic did not immediately respond to requests for comment.