OpenAI Acquires Promptfoo to Strengthen AI Security Tools

OpenAI is acquiring AI security platform Promptfoo to enhance testing, safety, and governance tools for enterprise AI systems. The technology will be integrated into OpenAI’s Frontier platform for AI coworkers.

By Maria Konash Published:
OpenAI Acquires Promptfoo to Strengthen AI Security Tools
OpenAI acquires Promptfoo to add AI security testing and evaluation tools to its Frontier platform for enterprise AI agents. Photo: fabio / Unsplash

OpenAI has announced plans to acquire Promptfoo, an AI security platform focused on identifying vulnerabilities in large language model applications during development. The company said Promptfoo’s technology will be integrated into OpenAI Frontier, its platform designed for building and operating AI coworkers in enterprise environments.

Promptfoo provides tools that help organizations evaluate, test, and secure AI systems before deployment. These capabilities are increasingly important as enterprises begin deploying AI agents into operational workflows that interact with sensitive data, internal systems, and external applications.

The acquisition aims to strengthen OpenAI’s ability to support enterprise customers that require structured approaches to evaluating agent behavior, identifying risks, and maintaining oversight over AI systems.

“Promptfoo brings deep engineering expertise in evaluating, securing, and testing AI systems at enterprise scale,” said Srinivas Narayanan, OpenAI’s chief technology officer for B2B applications. “Their work helps businesses deploy secure and reliable AI applications, and we’re excited to bring these capabilities directly into Frontier.”

Promptfoo was founded by Ian Webster and Michael D’Angelo and has developed a widely used open-source command-line interface and library for testing and red-teaming large language model applications. According to OpenAI, the platform is already used by more than 25 percent of Fortune 500 companies.

Security and Governance for AI Agents

OpenAI said Promptfoo’s technology will enable several new capabilities within the Frontier platform. Automated security testing and red-teaming tools will help enterprises identify risks such as prompt injection attacks, jailbreak attempts, data leakage, and misuse of connected tools.

The integration will also embed security testing directly into development workflows, allowing teams to identify vulnerabilities earlier in the development process. OpenAI said this approach will help organizations deploy AI agents with stronger safety and reliability controls.

Another key component involves oversight and compliance features. Frontier will include integrated reporting and traceability tools designed to help enterprises document testing procedures, monitor system changes, and meet regulatory governance requirements.

Promptfoo’s founders said the move will allow the platform to expand its capabilities as AI systems become more integrated with real-world data and business operations.

“We started Promptfoo because developers needed a practical way to secure AI systems,” said Ian Webster, co-founder and chief executive of Promptfoo. “As AI agents become more connected to real data and systems, securing and validating them is more challenging and important than ever.”

OpenAI said it plans to continue supporting Promptfoo’s open-source tools while expanding enterprise security capabilities through the Frontier platform. The acquisition reflects growing demand among organizations for robust testing and governance tools as AI agents move from experimentation into production environments.

AI & Machine Learning, Cybersecurity & Privacy, News, Startups & Investment
Related Articles

Anthropic Sues Pentagon Over Supply Chain Risk Designation

Anthropic has filed a lawsuit against the U.S. Department of Defense and other federal agencies over its designation as a supply chain risk. The company argues the move is unlawful and threatens its business relationships.

By Maria Konash Published:
Anthropic Sues Pentagon Over Supply Chain Risk Designation
Anthropic sues the Pentagon over its supply-chain risk designation. Photo: Tingey Injury Law Firm / Unsplash

Anthropic has filed a lawsuit against the U.S. Department of Defense and other federal agencies after the Trump administration labeled the artificial intelligence company a “supply chain risk.” The designation followed a breakdown in negotiations between the Pentagon and the developer of the Claude AI models over restrictions on military use of its technology.

In its legal filing, Anthropic described the government’s actions as “unprecedented and unlawful,” arguing that the designation and the directive requiring federal agencies to stop using its technology lack legal authority and proper due process.

“Seeking judicial review does not change our longstanding commitment to harnessing AI to protect our national security, but this is a necessary step to protect our business, our customers, and our partners,” an Anthropic spokesperson said in a statement.

The Pentagon declined to comment on the litigation. A White House spokesperson defended the administration’s position, stating that the government would not allow technology companies to dictate how military tools are used.

Dispute Over Military Use of AI

The conflict stems from negotiations to update the Pentagon’s contract with Anthropic. During the talks, the company asked the Defense Department to formally commit to two restrictions: that its AI systems would not be used for mass domestic surveillance of U.S. citizens and would not power fully autonomous weapons systems.

Defense officials rejected the conditions, insisting that the military must retain the ability to use AI for “all lawful purposes,” particularly in national security emergencies. The Pentagon has previously stated it does not intend to use AI for domestic surveillance or autonomous weapons but argued it could not accept restrictions imposed by a private company.

Following the breakdown in talks, the Trump administration ordered federal agencies and military contractors on February 27 to halt business with Anthropic. Defense Secretary Pete Hegseth also designated the company a supply chain risk, a classification typically applied to companies linked to foreign adversaries.

The designation limits Anthropic’s ability to work with companies that maintain contracts with the Department of Defense.

Claims of Constitutional Violations

Anthropic’s lawsuit alleges the government’s actions constitute retaliation against the company’s First Amendment protected speech. The filing also claims the administration exceeded its authority by directing federal agencies to cease using Anthropic’s technology without proper legal justification.

The company is seeking injunctive relief to prevent enforcement of the directive. According to the filing, the government’s actions place “hundreds of millions of dollars” in contracts at risk and could damage Anthropic’s reputation and commercial relationships.

Chief Executive Dario Amodei said the official designation letter suggests that contractors may still use Claude outside work directly tied to Pentagon contracts. The company has previously said it would challenge the classification in court, arguing that it sets a dangerous precedent for U.S. technology firms negotiating with government agencies.

Industry Impact and Public Reaction

The dispute has quickly escalated into one of the most significant confrontations between the U.S. government and an AI company over the limits of military technology deployment. Shortly after the administration’s directive, OpenAI reached a separate agreement with the Pentagon to deploy its models within defense infrastructure.

At the same time, Anthropic’s public profile has risen amid the conflict. The company’s Claude application recently overtook ChatGPT in Apple’s U.S. App Store rankings following the controversy, and Anthropic said more than one million new users are signing up for the platform each day as interest in its AI tools continues to grow.

AI & Machine Learning, News, Regulation & Policy
Related Articles

Microsoft Adds Anthropic AI to Copilot With New Cowork Tool

Microsoft is integrating Anthropic’s Claude models into Microsoft 365 Copilot and introducing a new Copilot Cowork tool for autonomous workflows. The move expands Microsoft’s AI partnerships as demand grows for agent-based productivity tools.

By Samantha Reed Edited by Maria Konash Published:
Microsoft Adds Anthropic AI to Copilot With New Cowork Tool
Microsoft integrates Anthropic’s Claude into Copilot and launches Copilot Cowork. Photo: Matthew Manuel / Unsplash

Microsoft is adding artificial intelligence models from Anthropic to its Microsoft 365 Copilot platform and introducing a new productivity feature called Copilot Cowork. The announcement reflects growing demand for AI agents capable of handling complex tasks across enterprise software environments.

Copilot Cowork is based on Anthropic’s Claude Cowork technology, which recently attracted attention in Silicon Valley for its ability to automate multi-step workflows. The tool can generate applications, build spreadsheets, and organize large datasets with limited human intervention.

Microsoft said Copilot Cowork will initially launch in testing and become available to early access users later this month. Pricing details were not disclosed, though the company said some functionality will be included in the existing Microsoft 365 Copilot subscription priced at $30 per user per month, with additional usage available separately.

The company is also making Anthropic’s Claude Sonnet models available within Microsoft 365 Copilot. Until now, the service relied primarily on models developed by OpenAI.

Enterprise Strategy and AI Partnerships

Microsoft is positioning the new tool as a secure enterprise-grade alternative for companies exploring AI agents but concerned about data protection and governance.

“We work only in a cloud environment and we work only on behalf of the user. So you know exactly what information it has access to,” said Jared Spataro, who leads Microsoft’s AI-at-Work initiatives.

According to Spataro, many organizations remain cautious about AI systems that operate locally without centralized oversight. Microsoft’s cloud-based approach aims to address those concerns by providing enterprise security controls and compliance tools.

The launch follows increased investor attention around agent-based AI products. Anthropic’s recent releases for Claude sparked speculation that AI agents could disrupt traditional software companies by automating tasks currently handled by specialized business applications. Those concerns contributed to volatility in software stocks earlier this year, including a decline of nearly 9 percent in Microsoft’s share price in February.

Shifting Dynamics in the AI Ecosystem

By integrating Anthropic’s models into Copilot, Microsoft is expanding its AI ecosystem beyond its long-standing collaboration with OpenAI. Analysts have increasingly scrutinized Microsoft’s reliance on OpenAI technology, which accounts for a substantial portion of its cloud-related AI backlog.

The new partnership allows Microsoft to diversify its model providers while continuing to expand the capabilities of Copilot as enterprises adopt generative AI tools across workplace applications.

The move also highlights intensifying competition among technology companies to provide AI-powered agents that can automate knowledge work, manage business workflows, and interact with enterprise software systems with minimal human supervision.

AI & Machine Learning, Enterprise Tech, News
Related Articles

Xiaomi Launches miclaw AI Agent Closed Beta in China

Xiaomi has launched a closed beta for miclaw, an AI agent powered by the MiMo large model that can control smartphones and smart home devices using natural language commands.

By Daniel Mercer Edited by Maria Konash Published:
Xiaomi Launches miclaw AI Agent Closed Beta in China
Xiaomi launches the miclaw AI agent beta in China, enabling natural-language control of smart devices. Photo: Xiaomi

Xiaomi has announced the limited closed beta release of miclaw, a new AI agent designed to control smartphones and connected smart home devices using natural language commands. The system is powered by Xiaomi’s MiMo large language model and represents the company’s latest step toward integrating AI-driven automation into its mobile ecosystem.

The invitation-only beta is currently available in China for select flagship devices, including the Xiaomi 17 Ultra Leica Edition, Xiaomi 17 Ultra, Xiaomi 17 Pro Max, Xiaomi 17 Pro, and the standard Xiaomi 17.

Unlike traditional voice assistants that respond to predefined commands, miclaw is designed to understand user intent and execute multi-step workflows across apps and system tools. The agent can access more than 50 system utilities and ecosystem services to complete tasks automatically.

Xiaomi said the system can also analyze user behavior and provide recommendations based on device data. For example, the AI agent may review subscription expenses and suggest potential savings opportunities, or adjust device settings based on contextual information such as calendar events.

AI Architecture and Multi-Step Automation

The underlying architecture uses an inference-execution loop that processes user input, selects relevant tools, and executes tasks while monitoring outcomes in real time. This process allows the AI to handle complex requests while maintaining responsiveness on the device.

A key feature of the system is its three-tier memory architecture. The AI retains important decision points and compresses interaction history to maintain context across longer workflows, supporting up to 20 consecutive steps within a single task.

Through integration with Xiaomi’s HyperConnect ecosystem and Mi Home protocols, miclaw can control a wide range of connected devices. Users can manage smart home equipment such as lighting, air conditioners, security systems, and robotic appliances using natural language instructions.

The system also supports developer integrations through the Model Context Protocol and an open software development kit. Applications can declare their capabilities to the AI agent, allowing it to dynamically discover and use third-party features.

Privacy and Edge Processing

Xiaomi said the AI system was designed with strict privacy safeguards. Most processing occurs through edge-cloud computing that keeps sensitive data on the device whenever possible.

The company also stated that personal interaction data from miclaw will not be used to train its AI models.

The development also aligns with Xiaomi’s broader technology strategy. The company has said it plans to release a new smartphone processor each year and introduce an AI assistant for international markets as it expands its chip design capabilities and global AI ecosystem.

AI & Machine Learning, Consumer Tech, News
Related Articles

OpenAI Launches ChatGPT for Excel With Financial Data Integrations

OpenAI has introduced ChatGPT for Excel in beta, enabling users to build and analyze spreadsheet models directly within workbooks. The release also adds financial data integrations from major providers such as FactSet and Dow Jones Factiva.

By Daniel Mercer Edited by Maria Konash Published:
OpenAI Launches ChatGPT for Excel With Financial Data Integrations
OpenAI launches ChatGPT for Excel with GPT-5.4 support and financial data integrations. Photo: OpenAI

OpenAI has launched ChatGPT for Excel in beta, introducing an add-in that embeds its AI assistant directly inside Microsoft Excel workbooks. The tool allows users to build spreadsheet models, update formulas, run scenario analysis, and generate insights using natural language prompts.

Powered by the new GPT-5.4 model, the add-in is designed to help analysts and finance professionals perform complex spreadsheet tasks faster while maintaining Excel’s native formulas and structure. Instead of manually building models or writing formulas, users can describe the task in plain language and have ChatGPT generate or modify spreadsheet logic directly in the workbook.

According to OpenAI, the feature enables users to analyze data across large spreadsheets, trace dependencies between cells and formulas, and identify errors or changes in model outputs. ChatGPT also links its explanations to specific spreadsheet cells, allowing users to verify how results were generated.

The system requires confirmation before editing workbooks and allows users to undo modifications, providing an additional layer of control for financial and analytical workflows.

OpenAI said the add-in aims to reduce time spent on manual tasks such as building models, reconciling spreadsheets, and debugging formulas. Analysts, accountants, and strategists can instead focus on interpretation and decision-making.

Early testing shows significant performance improvements for financial modeling tasks. On an internal investment banking benchmark measuring workflows such as building three-statement financial models, GPT-5.4 achieved an average score of 87.3% compared with 43.7% for earlier GPT-5 models.

Financial Data Integrations Expand ChatGPT’s Role in Research

Alongside the Excel add-in, OpenAI introduced new financial data integrations directly within ChatGPT. These connectors allow users to access market and company data from providers including Dow Jones Factiva, LSEG, Daloopa, S&P Global, and other financial data platforms, with FactSet integration expected soon.

The integrations enable users to combine proprietary datasets with ChatGPT’s reasoning capabilities for research tasks such as earnings analysis, valuation modeling, and investment due diligence.

ChatGPT can generate structured outputs including earnings summaries, credit analyses, and valuation snapshots while citing underlying data sources. Teams can also export generated reports to formats such as Microsoft Word or PDF for documentation and internal reporting.

OpenAI said the integrations are part of a broader ecosystem of applications built on the Model Context Protocol, which allows organizations to connect proprietary data sources and internal systems to ChatGPT workflows.

Enterprise Security and Governance

The company emphasized that the new capabilities are designed for enterprise environments, particularly in regulated industries such as financial services. ChatGPT Enterprise includes role-based access controls, single sign-on integration, audit logs, and compatibility with common data security tools.

Data transmitted through the system is encrypted both in transit and at rest, and OpenAI said customer data shared with ChatGPT Enterprise is not used to train its models by default.

ChatGPT for Excel is initially rolling out in beta to Business, Enterprise, Education, Teachers, Pro, and Plus users in the United States, Canada, and Australia. OpenAI also said support for Google Sheets is expected in a future release.

The launch reflects a growing push by AI developers to embed generative models directly into professional software tools. By combining GPT-5.4’s reasoning capabilities with spreadsheet workflows and financial datasets, OpenAI aims to streamline research, modeling, and analysis tasks across banking, asset management, and corporate finance.

AI & Machine Learning, Consumer Tech, Enterprise Tech, News
Related Articles

Everything You Should Know About OpenAI’s New GPT-5.4

OpenAI introduced GPT-5.4, a new frontier model combining advanced reasoning, coding, and agentic capabilities. The model brings improved tool use, computer interaction, and long-context performance across ChatGPT, the API, and Codex.

By Daniel Mercer Edited by Maria Konash Published:
Everything You Should Know About OpenAI’s New GPT-5.4
OpenAI launches GPT-5.4 with stronger reasoning, coding, and computer-use capabilities. Photo: OpenAI

OpenAI has released GPT-5.4, its latest frontier artificial intelligence model designed to combine reasoning, coding, and agentic workflows into a single system. The model is rolling out across ChatGPT, the developer API, and OpenAI’s coding platform Codex, which has been launched for Windows just recently.

The company also introduced GPT-5.4 Pro, a higher-performance variant intended for complex professional workloads requiring deeper reasoning and higher accuracy.

“GPT-5.4 brings together the best of our recent advances in reasoning, coding, and agentic workflows into a single frontier model,” OpenAI said in its announcement.

The company added that the model is designed to help users complete complex professional tasks “accurately, effectively, and efficiently—delivering what you asked for with less back and forth.”

GPT-5.4 builds on recent advances from earlier models, integrating the coding capabilities of GPT-5.3-Codex with improvements in reasoning, tool usage, and professional productivity tasks. The model is designed to handle real-world work involving spreadsheets, documents, presentations, and complex multi-step workflows.

In ChatGPT, the GPT-5.4 Thinking mode introduces a planning feature that outlines the model’s reasoning approach before completing a response. According to OpenAI, the feature allows users to “adjust course mid-response while it’s working” to produce outputs more closely aligned with their needs. The update also improves deep web research capabilities and long-context reasoning for complex queries.

Benchmark results show measurable improvements over earlier models. On GDPval, a benchmark testing professional knowledge work across 44 occupations, GPT-5.4 matched or exceeded industry professionals in 83% of comparisons, compared with 70.9% for GPT-5.2.

Computer Use and Agentic Workflows

GPT-5.4 introduces native computer-use capabilities, allowing AI agents to interact with software environments and operating systems. The model can operate applications using keyboard and mouse actions, interpret screenshots, and execute workflows across multiple programs.

OpenAI described the release as “the first general-purpose model we’ve released with native, state-of-the-art computer-use capabilities.” The company said the feature enables developers to build agents capable of carrying out complex workflows across applications and websites.

On the OSWorld-Verified benchmark, which evaluates desktop navigation and computer interaction, GPT-5.4 achieved a 75% success rate. This represents a significant improvement over the 47.3% score recorded by GPT-5.2 and exceeds average human performance in the benchmark.

The model also improves visual reasoning and document parsing capabilities. On the MMMU-Pro benchmark measuring multimodal reasoning, GPT-5.4 reached 81.2% accuracy without tools. The model also introduces higher-fidelity image input support, enabling analysis of images up to 10.24 million pixels.

Coding and Developer Tools

GPT-5.4 incorporates the coding strengths of GPT-5.3-Codex while improving performance on longer-running development workflows. On the SWE-Bench Pro benchmark, which measures software engineering performance, GPT-5.4 achieved a score of 57.7% while maintaining lower latency compared with earlier models.

OpenAI also improved how models interact with external tools through a feature called tool search. Instead of loading every tool definition into the prompt, the system retrieves them dynamically when needed.

“This approach dramatically reduces the number of tokens required for tool-heavy workflows and preserves the cache, making requests faster and cheaper,” the company said.

Internal tests showed the feature reduced token usage by roughly 47% while maintaining the same accuracy.

In Codex, developers can also use a new experimental capability called Playwright Interactive, which allows the model to visually debug web applications and test software as it is being generated. OpenAI said the feature helps developers automate browser interactions and validate application behavior during development.

Availability and Pricing

GPT-5.4 is rolling out gradually across OpenAI’s platforms. In ChatGPT, GPT-5.4 Thinking replaces GPT-5.2 Thinking for Plus, Team, and Pro subscribers, while GPT-5.4 Pro is available to Pro and Enterprise users.

Developers can access the model through the API using the gpt-5.4 and gpt-5.4-pro endpoints. Pricing starts at $2.50 per million input tokens and $15 per million output tokens, reflecting the model’s expanded capabilities.

OpenAI said the release represents a major step toward more capable autonomous systems. By combining reasoning, coding, computer interaction, and tool use, GPT-5.4 is designed to support “more reliable agents, faster developer workflows, and higher-quality outputs across ChatGPT, the API, and Codex.”

The release also aligns with OpenAI’s broader push to expand its developer ecosystem. The company is reportedly developing a new code hosting platform that could compete with Microsoft-owned GitHub, reflecting a strategy to integrate AI models, coding agents, and developer infrastructure into a unified platform.

AI & Machine Learning, Consumer Tech, Enterprise Tech, News
Related Articles