OpenAI deploys ChatGPT Enterprise as Google patches Antigravity vulnerability

Key Takeaways

• Hyatt deployed OpenAI's ChatGPT Enterprise, granting access to GPT 5.4 and Codex for finance, marketing, and operations teams.
• Researchers found a critical vulnerability in Google's Antigravity AI manager allowing sandbox escape via prompt injection, patched on February 28.
• A Codex vulnerability revealed how malicious AGENTS.md files can inject instructions into development environments using specific environment variables.
• B2B software contracts are shortening, with sub-1-year deals increasing from 4% in 2023 to 13% in 2026 as sales cycles compress.
• Hollywood accepts utility AI for efficiency but rejects generative AI that creates content or claims authorship.
• Controversy arose over Google's internal AI usage, with claims that DeepMind engineers use Anthropic's Claude while others use internal Gemini.
• Penn researchers launched the AIRFoundry hub to accelerate AI-driven RNA drug development and reduce testing costs.
• Robbie Williams sparked debate about AI mimicking artists, raising questions about intellectual property and creative ownership.
• Drake University students criticized inconsistent AI policies across courses, noting faculty also use AI-generated assignments.
• The Constructive Dialogue Institute is piloting AI chatbots to help polarized college students practice civil dialogue and perspective-taking.

Hyatt Deploys ChatGPT Enterprise for Global Workforce

Hyatt has deployed ChatGPT Enterprise to its global corporate and hotel workforce to improve productivity and guest experiences. Employees now have access to frontier AI capabilities like GPT 5.4, Codex, and other advanced models. Departments including finance, marketing, and operations will use these tools to streamline tasks and enhance service. Hyatt collaborated with OpenAI to provide live onboarding and training sessions to help teams integrate AI into their daily workflows. This initiative supports various functions such as accelerating financial reporting, scaling content creation, and personalizing guest interactions.

Hyatt Integrates OpenAI AI for Enhanced Operations

Hyatt Hotels Corporation is integrating OpenAI's advanced artificial intelligence into its daily operations by deploying ChatGPT Enterprise. The initiative grants employees access to frontier AI capabilities including models like GPT 5.4 and Codex. Finance teams will use the tool to accelerate closing cycles and improve reporting accuracy. Marketing and brand departments will scale content creation while maintaining consistency. Business development and engineering teams will leverage the AI for better investment research and faster digital platform development. This deployment aligns with Hyatt's broader strategy of building AI-powered experiences similar to its Hyatt app.

Google Antigravity AI Vulnerability Allows Sandbox Escape

Researchers at Pillar Security discovered a bug in Google's Antigravity AI agent manager that could allow attackers to escape the sandbox and gain remote code execution privileges. The vulnerability combined prompt injection with file-creation capabilities to bypass Secure Mode, Google's highest security setting for its agents. The bug was reported to Google on January 6 and patched on February 28, with Google awarding a bug bounty for the discovery. The exploit targeted a native system tool called find_by_name that the agent could execute before protections evaluated commands. This pattern of prompt injection through unvalidated input has also been found in other coding AI agents like Cursor.

Nvidia Blog Discusses AGENTS.md Injection Risks

A recent Codex vulnerability discovered by the Red Team highlights how malicious AGENTS.md files can execute undesired actions in agentic environments. The attack chain involved a Golang development project using a malicious library that overwrote the AGENTS.md file to inject instructions for the agent. The library selectively targeted Codex environments using the CODEX_PROXY_CERT environment variable to avoid execution in standard development environments. The injected instructions forced the agent to inject a five-minute delay into Golang main functions while remaining hidden in summaries and commit messages. This demonstrates how agent instruction files expand the attack surface beyond traditional prompt injection methods.

Penn Researchers Launch AI RNA Drug Development Hub

Penn researchers are working to accelerate the development of drug delivery systems through a new artificial intelligence-driven research hub called AIRFoundry. Chemical and biomolecular engineering professor Daeyeon Lee serves as the director and aims to democratize RNA technology across disciplines. The initiative seeks to reduce technical barriers that currently limit broader uses of RNA tools for drug synthesis and delivery. Graduate student Andrew Hanna helps design lipid nanoparticle formulation systems used to generate datasets for machine learning models. The goal is to speed up drug development by improving predictive systems and reducing trial-and-error testing through shared collaboration.

B2B Buyers Demand Shorter AI Software Contracts

Customers in the B2B software sector are increasingly asking for shorter contracts in the age of rapidly evolving AI technology. Sub-1-year contracts for new logo subscriptions have grown from 4% of deals in 2023 to 13% in 2026, while three-year deals have dropped from 28% to 23%. Sales cycles have also shortened from 25 weeks in H1 2025 to 19 weeks in H2 2025 as buyers want flexibility to adapt to changing market leaders. Many companies now use hybrid pricing models where 48% have this as their primary model, making long-term commitments harder to approve. Buyers view long-term contracts as risks in an AI world where the best solution can change in a matter of months.

Duke University Balances STEM and Humanities Funding

Duke University has split its investments between STEM growth and humanities expansion despite external funding favoring technology fields. While STEM fields, artificial intelligence, and quantum-related initiatives dominate philanthropic giving, hiring decisions and programming expansions seem to favor the humanities. In 2025, 50% of new hires were for humanities departments, followed by 31% for natural sciences and 19% for social sciences. The University launched new interdisciplinary technology programs and expanded AI research through grants and awards. Associate dean Owen Covington noted that cost-cutting affected departmental investment but the University continues to invest in areas central to its mission.

Robbie Williams Questions AI Boundaries in Music

British singer-songwriter Robbie Williams has sparked a debate about the use of artificial intelligence in the music industry and its impact on artists. Williams expressed concerns about AI knowing him better than he knows himself and questioned where the line sits regarding AI-generated content. He highlighted the ethical and legal implications for artists' creative processes and intellectual property in an era where AI can mimic voices and generate lyrics. The conversation has gained traction as companies develop algorithms to create music and write lyrics, raising questions about ownership and authenticity. Williams advocates for a serious conversation about protecting artists' rights as the industry navigates this new landscape.

Controversy Erupts Over Google AI Adoption Claims

A drama between software engineer Steve Yegge and Google heats up over claims about internal AI adoption disparities within the company. Yegge claimed there is a two-tier system where DeepMind engineers use Anthropic's Claude frequently while the rest of Google uses internal Gemini variants. Google AI boss Demis Hassabis and Google Cloud director Addy Osmani disputed these claims, calling them false and noting that over 40,000 Google software engineers use agentic coding weekly. Yegge argued that weekly use of a thin tool represents box-checking rather than true adoption and that volume of opens does not equal meaningful usage. The post sparked public backlash from Google workers ranging from top executives to lower-level engineers.

Hollywood Accepts Utility AI but Rejects Generative AI

Hollywood has become one of the loudest voices of resistance to generative AI while quietly accepting utility AI for efficiency gains. Filmmakers and editors draw a clear distinction between AI that increases efficiency and AI that creates content or claims authorship. Tools like TACHI AI are gaining acceptance because they automate technical aspects of editing such as assembling initial cuts without interfering with creative storytelling. Industry leaders argue that AI should support the creative process by giving time back to editors and directors rather than directing it. This shift aims to avoid ethical issues and legal problems while maintaining competitiveness in a fast-paced post-production environment.

Students Criticize Drake University AI Policy Inconsistencies

A student at Drake University argues that the university's double standards with AI usage are unfair and inconsistent across different courses. The student noticed that some classes require students to use AI for random tasks while others strictly forbid it, creating confusion about acceptable practices. The student pointed out that faculty members purchase and use AI-generated assignments, which contradicts the rules given to students. A study cited in the article suggests that faculty reliance on AI diminishes their inclination to critically evaluate outputs, potentially setting a bad example for students. The student calls for a consensus on AI usage or a commitment to using it only as a tool rather than an offload of cognitive function.

AI Chatbots Help Polarized College Students Dialogue

College students are more polarized than ever, and some institutions are using AI-powered platforms to foster civil dialogue and bridge divides. The Constructive Dialogue Institute is piloting an AI-enabled component that coaches students on actively listening and finding common ground amid fundamental disagreement. The AI chatbot presents hypothetical scenarios ranging from roommate disputes to debates about war in the Middle East and gives feedback on student responses. Students report that the tool creates a safe space to practice skills like perspective taking and asking constructive questions without the fear of messing up. Researchers suggest that AI's role in dialogue should be constrained and pedagogically focused to maximize benefits and minimize risks.

Sources

• OpenAI helps Hyatt advance AI among colleagues
• Hyatt Embraces ChatGPT Enterprise
• Vuln in Google’s Antigravity AI agent manager could escape sandbox, give attackers remote code execution
• Mitigating Indirect AGENTS.md Injection Attacks in Agentic Environments
• Penn researchers use AI to accelerate RNA drug development
• It’s Not Just You. Customers Are Asking for Shorter and Shorter Contracts in the Age of AI
• Duke’s investments split between STEM growth and humanities expansion
• Robbie Williams ‘Raises Questions About Where the Line Sits’ With AI Use
• Drama between software engineer and Google heats up
• AI In Hollywood: Saying Yes To Utility AI – And No To Generative AI
• The University’s double standards with AI aren’t fair
• College Students Are More Polarized Than Ever. Can AI Help?