openai launches google while anthropic expands its platform

OpenAI is aggressively expanding its AI capabilities into business workflows with a suite of new agents designed to automate repetitive tasks. The company launched Tally, an agent that automatically generates weekly product metrics reports by reading Google Sheets and calculating statistics for the past seven days. This tool runs every Friday at noon to deliver consistent insights without manual effort.

Complementing Tally, OpenAI introduced Trove to automate third-party vendor risk assessments. This agent evaluates companies against risk criteria stored in Google Drive, producing detailed reports on security, financial, and operational risks to speed up due diligence. Meanwhile, the Spark feature allows users to build custom sales agents that research leads, grade their quality, and draft outreach emails via Gmail.

Workspace Agents are now available for teams, built on the Codex model to handle multi-step tasks within organizational security rules. These agents can monitor Slack, answer questions, and triage issues by connecting to tools like Linear. OpenAI's sales team is already using an agent to streamline lead qualification, and the feature is in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans.

Security concerns are rising alongside AI adoption. Silverfort and SentinelOne partnered to address identity security challenges, combining identity discovery with AI-powered detection to protect human users and autonomous agents. ServiceNow acquired Armis to extend AI-powered cyber risk visibility across IT, OT, IoT, and medical devices, creating a Context Engine for automatic risk prioritization. Additionally, exposed OpenClaw control panels on 28,663 unique IP addresses highlight the risks of giving AI agents full access to systems.

In the robotics sector, Sony's AI-developed robot Ace defeated elite table tennis professionals in a competitive match. The robot, equipped with an eight-jointed arm and nine camera eyes, won three out of five matches. Ace learned to play through 3,000 hours of computer simulations and reinforcement learning rather than hand programming, marking a milestone for real-world competitive robotics.

Legal and ethical questions are also surfacing. WIRED experts will host a livestream discussing the ongoing trial between Elon Musk and Sam Altman regarding OpenAI's founding mission. Separately, an undergraduate student leaked the source code for Anthropic's Claude Code AI agent, raising copyright questions about whether rewriting the code transforms it into a new work. Over 100,000 people engaged with the rewritten code within hours.

Industry leaders are adapting strategies for the AI era. Visa CMO Frank Cooper III advises businesses to structure data so machines can find and trust their products, creating structured catalogs and standardized schemas. TrendMicro is building company memory with AWS, integrating Amazon Neptune and Amazon Bedrock to manage conversational memory with human-in-the-loop feedback. Data security remains a top priority, with many CIOs allocating four times more budget to data infrastructure than to AI due to vulnerabilities in multi-vendor supply chains.

Key Takeaways

OpenAI launched Tally, an agent that automates weekly product metrics reports by reading Google Sheets and calculating statistics.
OpenAI introduced Trove to automate third-party vendor risk assessments using web search and Google Drive integration.
OpenAI's Spark feature enables users to build custom sales agents that research leads and draft emails via Gmail.
Workspace Agents, built on the Codex model, are now available for teams to handle multi-step tasks within organizational security rules.
Silverfort and SentinelOne partnered to address identity security challenges caused by increasing AI adoption in enterprises.
ServiceNow acquired Armis to extend AI-powered cyber risk visibility across IT, OT, IoT, and medical devices.
Security researchers found 28,663 unique IP addresses hosting exposed OpenClaw control panels vulnerable to remote code execution.
Sony's robot Ace defeated elite table tennis professionals using reinforcement learning and an eight-jointed arm.
An undergraduate student leaked the source code for Anthropic's Claude Code AI agent, raising copyright ownership questions.
Visa CMO Frank Cooper III advises businesses to structure data with consistent metadata to become knowledge sources for AI agents.

OpenAI launches Tally agent for weekly metrics

OpenAI released a new agent named Tally that automates the creation of weekly product metrics reports. The agent reads data from Google Sheets, calculates statistics for the last seven days, and generates clear updates. It integrates with Google Drive to access files and uses custom skills to standardize how metrics are processed. The system runs automatically every Friday at 12:00 PM to deliver consistent insights without manual effort.

OpenAI unveils Trove for vendor risk analysis

OpenAI introduced Trove, a custom AI agent designed to automate third-party risk assessments for vendors. The agent uses web search and a specialized research skill to evaluate companies against a risk criteria spreadsheet stored in Google Drive. It produces a detailed Google Doc report covering security, financial, and operational risks. This tool helps businesses speed up due diligence by replacing slow manual processes with consistent AI-driven analysis.

OpenAI Spark automates sales lead tasks

OpenAI's Spark feature allows users to build custom agents, including a sales agent that handles lead research and outreach. The agent uses web search to find new leads, grades their quality, and drafts emails via Gmail. It also schedules follow-up reminders in Google Calendar to ensure timely communication. In a demonstration, the agent automatically responded to a lead inquiry about OpenAI products by researching and sending a tailored message.

ChatGPT Workspace Agents boost productivity

OpenAI launched Workspace Agents, which let users create personalized AI assistants for specific professional tasks. These agents can monitor channels like Slack, answer questions, and triage issues by connecting to tools such as Linear. Examples include automating lead outreach, managing month-end closes, and tracking key performance indicators. The agents operate 24/7 to handle repetitive work and improve team efficiency across various business functions.

OpenAI expands AI into business workflows

OpenAI is expanding ChatGPT capabilities with workspace agents that automate repeatable tasks and connect to existing business tools. Unlike standard APIs, these probabilistic agents interpret context and make decisions within defined parameters. They are best suited for structured, time-based, or event-driven tasks that require interaction with specific systems. Users can build these agents by defining objectives in plain language and selecting appropriate tools and triggers.

OpenAI rolls out team-focused Workspace Agents

OpenAI is rolling out Workspace Agents for ChatGPT to transform team collaboration and automate complex workflows. Built on the Codex model, these agents handle multi-step tasks while operating within organizational security and governance rules. OpenAI's sales team is already using an agent to streamline lead qualification and draft follow-up emails. The feature is currently in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans.

New ChatGPT Workspace Agents for teams

OpenAI introduced Workspace Agents in ChatGPT, allowing teams to create shared agents for complex tasks and long-running workflows. Powered by Codex, these agents can prepare reports, write code, and respond to messages while running in the cloud. They are designed to be shared within organizations, enabling teams to build agents once and use them together in ChatGPT or Slack. The agents operate within permissions and controls set by the organization to ensure safety.

Slate Agent streamlines software reviews

OpenAI launched Slate, an AI agent designed to streamline software reviews and reduce tool sprawl in businesses. The agent acts as a virtual reviewer for IT and sourcing departments, checking requests against an approved list of tools. It uses web search to compare requested software with existing options and provides data-driven recommendations. Slate integrates with Slack, Google Sheets, and Atlassian to manage requests efficiently.

Silverfort and SentinelOne partner on security

Silverfort and SentinelOne have partnered to address identity security challenges caused by increasing AI adoption in enterprises. The collaboration combines Silverfort's identity discovery with SentinelOne's AI-powered detection and response capabilities. The goal is to expand runtime protection across human users, non-human entities, and autonomous AI agents. The unified solution aims to detect and contain incidents in real time across cloud and on-premises environments.

ServiceNow acquires Armis for cyber visibility

ServiceNow completed its acquisition of Armis to extend AI-powered cyber risk visibility across OT, IoT, and other connected assets. The deal combines Armis's real-time asset intelligence with Veza's Access Graph to create a comprehensive Context Engine. This integration allows for automatic risk prioritization and autonomous remediation with full audit trails. The solution covers IT, OT, IoT, medical devices, code, and cloud environments.

Sony robot Ace defeats table tennis pros

Sony AI developed a robot named Ace that defeated elite table tennis players in a competitive match. The robot won three out of five matches against professionals using an eight-jointed arm and nine camera eyes to track the ball. Ace learned to play through 3,000 hours of computer simulations and reinforcement learning rather than hand programming. The achievement marks a milestone for robotics in real-world competitive sports.

Sony robot Ace challenges human table tennis

Sony built a robotic arm called Ace that poses a tough challenge to elite human players in table tennis. The robot uses nine camera eyes to follow the ball's logo and measure spin with high precision. Researchers used reinforcement learning to teach Ace to play, as the system cannot be programmed manually for this sport. The robot played on an Olympic-sized court in Tokyo under official rules to ensure fair competition.

WIRED livestreams Musk v. Altman trial

WIRED experts will host a subscriber-only livestream on May 8 to discuss the ongoing legal case between Elon Musk and Sam Altman. The event will allow attendees to submit questions about the trial regarding OpenAI's founding mission. A jury will soon determine whether OpenAI has strayed from its goal to ensure AGI benefits humanity. Subscribers can watch the live stream or access a replay afterward.

2026 Enterprise AI Data Security priorities

Data security has become a central priority for enterprises accelerating AI adoption in 2026. Leaders face challenges like unclear data residency, weak input controls, and vulnerabilities across multi-vendor supply chains. Many CIOs allocate four times more budget to data infrastructure than to AI due to these concerns. Security must be embedded from day one with governance, encryption, and infrastructure controls to ensure safe adoption.

Anthropic code leak raises copyright questions

An undergraduate student leaked the source code for Anthropic's Claude Code AI agent and posted it online. He then used AI assistants to rewrite the code in a different programming language and shared the new version. Over 100,000 people liked or linked to the rewritten code within hours. The student questioned whether rewriting the code transformed it into a new work that Anthropic could not claim ownership over.

Visa CMO advises selling to AI agents

Visa CMO Frank Cooper III argues that AI agents are becoming a new customer segment for businesses. Companies must structure their data so machines can find and trust their products. This involves creating structured product catalogs, consistent metadata, and standardized schemas for machine interpretation. Businesses should also become knowledge sources that AI systems can reason from rather than just product pages.

TrendMicro builds company memory with AWS

TrendMicro enhanced its AI chatbot service by integrating Amazon Neptune, Amazon OpenSearch, and Amazon Bedrock. Amazon Neptune stores a company-specific knowledge graph for precise retrieval, while Mem0 manages short-term and long-term conversational memory. The system uses a closed-loop process to continuously refresh memories and ensure the knowledge graph remains consistent. Users can approve or reject memory mappings through a human-in-the-loop feedback mechanism.

FSU faculty showcase AI and VR teaching

Florida State University faculty showcased innovative teaching methods including AI-driven coursework and virtual reality simulations. The third annual Provost's Showcase of Scholarly Teaching featured 31 poster presentations and 18 roundtable discussions. Educators highlighted strategies like gamifying persuasion through Dungeons & Dragons and using AI in introductory programming. The event aimed to improve student learning through research-based, student-centered approaches.

OpenClaw vulnerabilities expose thousands of systems

Security researchers found that 28,663 unique IP addresses host exposed OpenClaw control panels accessible from the internet. Approximately 63% of these deployments are vulnerable to remote code execution, allowing attackers to take over machines without user interaction. The exposed instances are heavily concentrated in major cloud and hosting providers with repeatable insecure patterns. Researchers warn that giving AI agents full access creates significant security risks.