Meta has quietly removed the word 'train' from its commercial terms of service ahead of an October 30, 2026 effective date. The original language appeared to allow Meta to use automated tools to scrape content from websites linked to business customers for AI model training. Legal experts flagged the terms as potentially illegal and a way to sidestep licensing payments to news organizations. Meta has not publicly explained the change but pledged greater transparency about data collection practices.
The News/Media Alliance called the updated terms outrageous, while Meta reportedly plans to let businesses control crawler access through robots.txt files. The move follows intense scrutiny over how tech firms use published content to build AI capabilities.
In separate AI news, Aikido Security launched Noon on October 20, 2026, a platform designed to shield employees from AI-powered social engineering attacks. Noon combines email protection, training simulations, device security, and AI governance. It connects to Microsoft 365, Google Workspace, Slack, and Microsoft Teams. The platform draws on two September 2026 acquisitions: Cyberly, which served over 300,000 employees across 200 organizations, and Outkept, which reached more than 1,000 organizations and 250,000 users.
Firmable, an AI-native B2B sales platform, rolled out Tech Intelligence, a tool that reveals the software, cloud, and AI systems used by prospect companies. It uses four detection methods tied to verified company and contact data. A Firmable survey found 71% of US sales teams struggle most with lacking timing and context on prospects. Tech Intelligence is available as a paid add-on and through Firmable's MCP integrations in Claude, ChatGPT, and Copilot.
The Federal Reserve's October Beige Book reported that firms in five of its twelve districts cut hiring after investing in AI. Some companies reduced staff through attrition or delayed new hires as AI improved productivity. Most districts saw little overall employment change, though wage growth varied. Labor shortages continued for specialized and technical roles.
Vectorial raised $3 million to expand its SAPIENS model, which simulates thousands of users to test products before launch. The company claims SAPIENS predicts user behavior 30% more accurately than general-purpose LLMs and outperforms models like Be.FM and Centaur. It has modeled over 2.5 million real people and run more than 30,000 simulations. Customers include MIDI Health, GoodRx, Suki AI, Boat, and Concrete Software.
Gigamon introduced the GigaVUE-GSA110 appliance on October 20, 2026, extending its Deep Observability Pipeline for hybrid AI environments. The device handles Application Metadata Intelligence across up to 400 Gbps of traffic and optimized packet delivery for up to 800 Gbps. Gartner predicts that by 2028, over 50% of enterprises will use public cloud selectively for bursty workloads while keeping performance-sensitive AI tasks in private settings.
Elice, a South Korean data center firm, plunged 32% on its IPO debut in Seoul, the worst opening for a sizable IPO in the country. The drop renewed concerns about heavy AI infrastructure spending. JP Morgan strategist Stephanie Aliaga pointed to a gap between roughly $7 trillion in expected AI spending through 2030 and about $3.6 trillion in projected revenue. Separately, Nscale faces investor pushback over its $30 billion IPO target, with investors signaling $16 billion to $19 billion would be more realistic. Microsoft had been expected to anchor Nscale's biggest site in West Virginia, but that deal did not materialize.
A study by IANS and Artico Search found that 65% of large organizations with revenue above $10 billion now have a head of AI security role. About 18% made it a dedicated position, while 47% combined it with another leadership title. The role is far less common at smaller firms, with only 9% of $1 billion to $10 billion companies and 6% of sub-$1 billion firms having the position. Experience drives pay differences, with a gap of up to $50,000 between early-career staff and those with over 16 years in the field.
A new research paper challenged how visual token pruning in vision language models is evaluated. The authors argued that relying on overall task accuracy and token count fails to show whether pruning truly selects important visual tokens or delivers real speed gains. They introduced PAA-Bench, Random-Pass@k, and Preservation-Constrained Speedup as fairer testing methods. Experiments across eight approaches, three model backbones, and nine benchmarks showed that higher token compression does not always lead to faster performance.
Key Takeaways
- Meta removed 'train' from its commercial terms ahead of an October 30, 2026 effective date, narrowing permitted data uses after backlash from legal experts and the News/Media Alliance.
- Aikido Security launched Noon, an AI social engineering defense platform integrating email protection, simulations, and governance, with connections to Microsoft 365, Google Workspace, Slack, and Teams.
- Firmable's Tech Intelligence tool reveals prospect technology stacks and is available through MCP in Claude, ChatGPT, and Copilot.
- The Federal Reserve's Beige Book reports firms in five districts cut hiring after investing in AI, though most districts saw little overall employment change.
- Vectorial raised $3 million for its SAPIENS model, which it says predicts user behavior 30% more accurately than general-purpose LLMs and has modeled over 2.5 million real people.
- Gigamon's new GigaVUE-GSA110 appliance supports up to 400 Gbps of Application Metadata Intelligence and up to 800 Gbps of optimized packet delivery for hybrid AI environments.
- Elice dropped 32% on its IPO debut in Seoul, highlighting investor anxiety over AI infrastructure spending and a projected $7 trillion spend versus $3.6 trillion in revenue through 2030.
- Nscale faces investor resistance over its $30 billion IPO target, with investors pushing for a $16 billion to $19 billion valuation; Microsoft did not materialize as the anchor customer for its West Virginia site.
- 65% of large organizations with over $10 billion in revenue now have a head of AI security, though the role is rare at smaller firms and pay gaps can reach $50,000 based on experience.
- A new paper introduces PAA-Bench, Random-Pass@k, and Preservation-Constrained Speedup to more fairly evaluate visual token pruning in vision language models, showing higher compression does not always mean faster performance.
Meta removes 'train' from AI terms before October 30 deadline
Meta quietly removed the word 'train' from a section of its business terms of service. The original language allowed Meta to use automated tools to crawl websites for AI training. After coverage by outlets like Engadget, Meta narrowed the list of permitted purposes. The updated terms are set to take effect on October 30, 2026.
Meta drops 'train' word from commercial terms after backlash
Meta has removed the word 'train' from its commercial terms of service taking effect on October 30. Legal experts warned the terms appeared to allow Meta to scrape websites linked on business pages. Experts suggested Meta could use the policy to avoid paying licensing fees to news organizations. Meta has not commented on the change but promised more transparency about data collection.
Meta's new terms permit AI training on business customer content
Meta plans to implement new commercial terms on October 30, 2026. The terms allow Meta to use automated tools like web scrapers to collect content from websites linked to business customers. This content can be used to train and improve Meta's AI models. The News/Media Alliance called the terms outrageous and probably illegal. Meta reportedly plans to let businesses control crawler access through robots.txt files.
Aikido Security launches Noon platform against AI social engineering
Aikido Security launched Noon on October 20, 2026 to protect employees from AI-powered social engineering. Noon is built on two September 2026 acquisitions: Cyberly, which brought AI attack simulations, and Outkept, which brought email security training. The platform combines email protection, simulations, device protection, and AI governance into one tool. Noon connects to Microsoft 365, Google Workspace, Slack, and Microsoft Teams.
Aikido Security launches Noon to protect workers from AI attacks
Aikido Security launched Noon on October 20, 2026 to address AI-driven social engineering threats. The platform draws on acquisitions of Cyberly and Outkept completed in September 2026. Cyberly previously protected over 300,000 employees across 200 organizations. Outkept served more than 1,000 organizations and 250,000 users. Noon offers email protection, training simulations, device protection, and AI governance in one platform.
Fed report says AI investment leads to hiring cuts in five districts
The Federal Reserve's October Beige Book reports that firms in five of its twelve districts reduced hiring after investing in AI. Some companies cut jobs through attrition or delayed new hires because AI improved productivity. Most districts still saw little overall change in employment. Wage growth varied, with five districts reporting moderate growth and others reporting modest or slight increases. Labor shortages persisted for specialized and technical workers.
New study challenges how VLM token pruning is measured
A new paper argues that current ways of evaluating visual token pruning in vision language models are flawed. The study says relying on overall task accuracy and token count does not show whether pruning truly selects important visual tokens or delivers real speed improvements. The researchers introduced PAA-Bench, Random-Pass@k, and Preservation-Constrained Speedup to test pruning more fairly. Experiments across eight methods, three model backbones, and nine benchmarks showed that higher token compression does not always mean faster performance.
Firmable launches Tech Intelligence to reveal prospect technology stacks
Firmable, an AI-native B2B sales platform, launched Tech Intelligence, a tool that shows sales teams what software, cloud, and AI systems a prospect company uses. It combines four detection methods and links findings to verified company and contact data. A Firmable survey found 71% of US sales teams said lack of timing and context on prospects was their biggest problem. The tool is available as a paid add-on and through Firmable's MCP in Claude, ChatGPT, and Copilot.
Vectorial raises $3M to test products with simulated users
Vectorial raised $3 million to expand its platform that simulates thousands of users to test products before launch. Its SAPIENS model is said to predict user behavior 30% more accurately than general-purpose LLMs and outperforms models like Be.FM and Centaur. The company has modeled over 2.5 million real people and run more than 30,000 simulations. Customers include MIDI Health, GoodRx, Suki AI, Boat, and Concrete Software, and the funds will support SAPIENS 2.0 and new behavioral datasets.
Head of AI security roles surge in large organizations
A new study by IANS and Artico Search found that 65% of large organizations with revenue over $10 billion have established a head of AI security role. About 18% made it a dedicated position, while 47% combined it with another leadership role. The role is less common in smaller firms, with only 9% of $1 billion to $10 billion companies and 6% of sub-$1 billion firms having the position. Experience was the biggest factor in pay differences, with a gap of up to $50,000 between early-career staff and those with over 16 years of experience.
Elice plunges 32% in worst IPO debut, sparking AI funding fears
Elice, a South Korean data center firm, dropped 32% on its first day of trading in Seoul, marking the worst opening for a sizable IPO in the country. Bloomberg Technology reported that the fall renews worries about heavy spending on AI infrastructure. JP Morgan strategist Stephanie Aliaga noted a gap between roughly $7 trillion in expected AI spending through 2030 and about $3.6 trillion in expected revenue, pointing to possible write-downs. Investors are also pushing nScale toward a $16 billion to $19 billion valuation, well below management's $30 billion target.
Gigamon launches new appliance for AI network and security operations
Gigamon announced the GigaVUE-GSA110 network appliance on October 20, 2026, extending its Deep Observability Pipeline with distributed GigaSMART processing. The appliance supports Application Metadata Intelligence across up to 400 Gbps of network traffic and optimized packet delivery for up to 800 Gbps. It is designed for Hybrid AI environments where organizations use frontier, open-weight, specialized, and private models across public cloud, private cloud, and on-premises infrastructure. Gartner predicts that by 2028, more than 50 percent of enterprises will use public cloud selectively for bursty workloads while keeping performance-sensitive AI workloads in private environments.
Nscale faces investor pushback on $30 billion IPO target
Nscale is encountering investor resistance over its IPO valuation target of about $30 billion, with investors signaling that $16 billion to $19 billion would be needed to get the deal through. The company is a neo-cloud firm trying to build data centers faster than hyperscalers and rent capacity back to them, but it faces problems including power constraints, labor issues, and potential delays. Microsoft was expected to be the major customer for Nscale's biggest site in West Virginia but that did not materialize. Bloomberg Podcasts said the central question is whether long-term demand for AI computing will spread widely enough to justify the capacity being built.
Sources
- Meta Axes 'Train' From AI Terms Ahead of Oct. 30
- Meta drops the word "train" from its commercial TOS after AI backlash
- Meta's new terms allow AI training on business customer content
- Aikido Security launches Noon to protect employees from AI-powered social engineering
- Aikido Security launches Noon to protect employees from AI-powered social engineering
- Fed: firms in 5 of 12 districts curb hiring after AI investment
- Are We Evaluating VLM Token Pruning Right? Looking Beyond Accuracy and Token Count
- Firmable launches Tech Intelligence, showing B2B sales teams what their prospects actually run and who to call about it
- Vectorial raises $3M to simulate users before launch
- Head of AI Security Roles Surge in Large Organizations
- Elice Crashes 32% and AI Funding Fears Return
- Gigamon Extends the Deep Observability Pipeline for AI-Enabled Network and Security Operations with New GigaSMART Network Appliance
- Nscale Hit With Pushback on $30 Billion IPO Target
Comments
Please log in to post a comment.