Google Offers Buyouts in Five Countries Amid AI Spending Pressure

Key Takeaways

• Google expanded voluntary buyouts to five countries (Canada, France, Germany, Japan, Switzerland) amid heavy AI spending pressuring Alphabet stock.
• Visa found AI shopping models fail 42.8 percent of test runs, with 54 percent of failed purchases still completing transactions anyway.
• An EU court adviser said AI training on copyrighted works may be legal if developers avoid pirate sources and respect paywalls.
• The White House withheld 60 pages of AI security review records, citing national security concerns after a FOIA lawsuit.
• Microsoft announced Dynamic Permissions and Execution Containers for AI agent security at Ignite, running November 17-20, 2026.
• Merge uses GPT-6 Astra, ChatGPT and Codex from OpenAI, cutting executive dinner prep from hours to about 30 minutes per event.
• Researchers reduced benchmarking error for agentic systems from 0.13 to 0.08 on SWE-bench Verified using cached trace summaries.
• JFrog launched AI security tools with forecast revenue of US$981.6 million and earnings of US$121.2 million by 2029.
• Fortinet launched FortiGuard Inline AI Security Service and FortiLink SASE Service targeting AI-specific risks like prompts and data leakage.
• Firmly raised 15 million dollars in Series A funding, with early tests showing up to 18% conversion rate improvements for merchants using AI shopping channels.

Google expands exit program to five countries amid AI spending

Google offered voluntary buyouts to employees in Canada, France, Germany, Japan and Switzerland starting in June 2026. The program gives workers severance if they are uncomfortable with Google's future direction. Google Cloud and DeepMind employees were excluded from the offers. The move comes as the company faces pressure from heavy spending on artificial intelligence. Alphabet's stock has worried investors over these AI costs.

Google expands voluntary buyouts to 5 countries as AI spending grows

Google expanded its voluntary buyout program to workers in five countries including Canada, France, Germany, Japan and Switzerland. A Google spokesperson confirmed select employees were offered packages in June 2026. The buyouts come as heavy AI spending weighs on Alphabet stock. Microsoft also offered buyouts to 60,000 employees across several divisions last week. Both companies face financial pressure from massive investments in AI infrastructure and data centers.

Visa finds AI shopping models fail 43 percent of test runs

Visa tested 22 AI models from 12 labs on shopping tasks like buying a TV, laptop or personal care appliance. The overall pass rate was 57.2 percent, with the best models passing 82.2 percent and the worst passing only 28.9 percent. A concerning finding was that 54 percent of failed runs still ended in a purchase. The AI often bought items that did not meet customer requirements. Visa created this benchmark to measure trust in AI-powered shopping.

Visa tests whether AI models truly understand shopper needs

Visa tested AI models on 15 shopping challenges to see if they can identify what a consumer wants. Each model got three attempts per scenario, but all three attempts passed only 34.2 percent of the time. The worst result was that 54 percent of failed runs still led to a purchase. A Visa survey found only 23 percent of U.S. consumers trust AI to handle payment transactions. Visa plans to expand testing to more product categories in the future.

EU court adviser says AI training on copyrighted works may be legal

An EU court adviser said training AI on copyrighted books and articles should be lawful under three conditions. Developers must avoid pirate sources, respect paywalls and bot blocks, and show they tried to stop the model from copying specific works. The opinion was issued by Advocate General Maciej Szpunar on November 3, 2026. It concerns a Hungarian news website's claim against Google's Gemini chatbot. The adviser also suggested the court declare two questions about model training inadmissible.

White House withholds AI security review records

The White House refused to release 60 pages of records about its AI system security review process. The Protect Democracy Project sued the government under the Freedom of Information Act after an agreement to receive documents by October 30 was not honored. The review system examines new AI tools before public release, but officials have not explained how it operates publicly. The White House cited national security concerns for withholding the documents.

Microsoft Security updates cover agent permissions and containers

Microsoft announced new security features for November 2026 at Ignite, running November 17 to 20. Dynamic Permissions gives AI agents task-scoped authority that adjusts as risk changes, while Microsoft Execution Containers provide an operating system-enforced containment layer for agents on Windows. Defender can now discover over 50 supported local AI agents and their configured model context protocols across Windows and macOS. The enhanced agent for Defender for SQL Servers is generally available in Azure Government cloud.

Fortinet launches two new security services

Fortinet introduced the FortiGuard Inline AI Security Service and the FortiLink SASE Service. The FortiGuard service provides AI-specific inspection and guardrails for enterprises using generative AI tools. FortiLink SASE offers agentless, cloud-delivered security for distributed locations without on-site firewalls. The launch targets risks around prompts, data leakage, and AI tool use while building on Fortinet's existing firewall and SASE base.

Merge uses OpenAI technology for product and business tools

Merge uses GPT-6 Astra, ChatGPT, and Codex for product reviews, marketing, and business development planning. CEO Shensi Ding uses Codex for code reviews and GPT-6 Astra for product testing. Merge reported that preparing guest lists and outreach for executive dinners dropped from dozens of hours to about 30 minutes per event in October 2026. The company's tool calls through Merge succeed 99.9 percent of the time compared to about 60 percent for public MCP servers.

New method improves agent evaluation efficiency with cached traces

Researchers proposed a method for efficient benchmarking of agentic systems using structured trace summaries extracted by an LLM and embedded with a text encoder. The approach was tested on SWE-bench Verified and Terminal-Bench 2.0, reducing mean absolute error from 0.13 to 0.08 and from 0.14 to 0.12 respectively. The method constructs low-dimensional system representations and predicts new system scores from their position among reference systems. It requires only a single random executed task to achieve significant improvements.

JFrog Launches AI Security Tools to Strengthen Software Supply Chain

JFrog introduced new AI SEC and DevGovOps tools to improve software supply chain security and governance. The tools aim to integrate AI focused security into daily development workflows. Investors are watching whether these features increase security usage on existing products like Artifactory, Curation and Xray. Analysts forecast revenue could reach US$981.6 million and earnings of US$121.2 million by 2029 with steady growth.

Firmly Raises 15 Million Dollars to Connect Merchants with AI Shopping

Firmly raised 15 million dollars in a Series A funding round to help merchants connect with AI shopping channels. Its platform lets merchants link to AI agents without writing new code for each channel. Mastercard named Firmly a partner in September. Early tests of its Ask Agent tool showed conversion rate improvements up to 18% and average order value increases up to 23%.

Sources

• Google Expands Exit Program in Five Countries Amid Heavy AI Investment
• Google expands voluntary buyouts to 5 countries as AI spending fuels cost cuts
• Visa Finds AI Models Fail 43% of Shopping Test Runs
• Does AI Truly Understand What Shoppers Want? Visa Puts It To The Test
• EU court adviser in Google case backs AI training on 3 conditions
• White House Withholds Records on Its AI Model Security Review Process
• ​​​​What’s new in Microsoft Security: November 2026
• Fortinet (FTNT) Unveiled Two New Security Services
• Merge builds product and business tools with OpenAI technology
• Task-Efficient Agent Evaluation Using Cached Traces
• How AI Security Launches Will Impact JFrog (FROG) Investors
• Firmly Lands $15 Million as Investors Back the Merchant Side of Agentic Commerce