LinkedIn, YouTube, ZeroDrift and Lookout unveil AI tools for trust, content and security

LinkedIn said it will let members vouch for colleagues and classmates. The feature is not an endorsement but a confirmation that someone had the experience listed on their profile. To reduce abuse, the person vouching must have a verified profile, must have been connected to the member for at least a year, and must use two-factor authentication. LinkedIn also gave verified company page admins the ability to remove accounts that falsely claim employment from their page and associated search results. It is testing a setting that would require workplace verification, such as through a work email address, for members who want to associate with a page. Removing a profile from a company page will not change the member’s own profile, but the company link will no longer be clickable and the work experience will not display as verified. The feature focuses on full-time employees, not contractors or field workers.

LinkedIn Vice President of Product Oscar Rodriguez told TechCrunch that faking credibility has never been cheaper or easier, while showcasing credibility has never mattered more. LinkedIn said its existing verification tools have been used by 115 million users and more than 700,000 companies. Rodriguez said verified members receive, on average, 50 percent more post views and 90 percent more impressions than non-verified profiles. Truecaller and PeerSpot are joining Adobe and UserTesting as verification partners; Truecaller members will be able to display their LinkedIn verified identity to call recipients.

At its Made On YouTube event, YouTube announced updates to its AI-powered creator suite. An AI agent will work in the background to optimize a creator’s channel, monitoring the back catalog for videos that have become relevant or started trending and suggesting changes to thumbnails or titles. YouTube said the agent can also assemble brand pitches using demographic and audience data. Ask Studio lets creators share draft videos with AI for personalized feedback, alternative titles and thumbnails, channel analytics and advice. Dynamic Thumbnails lets creators upload up to three images per video, with YouTube assigning the best one to different audience segments. YouTube plans to expand A/B testing for Shorts to three different cuts from 2027 and is testing three versions of videos with small audience segments to find the highest watch time; creators can make the winner permanent or let YouTube do so after seven days. Amjad Hanif, YouTube’s vice president of creator products, said the system ensures the versions are not dramatically different.

YouTube declined to share data showing whether the tools improve metrics for creators who use them, saying they save creators time. Hanif said the time saved goes back into creativity. He distinguished between using AI for efficiency and tools doing the actual job of a creator, saying script writing, shooting scenes and wholesale content creation are the kinds of AI use that may cross a line for creators and viewers.

YouTube also announced features beyond creator tools. YouTube Music will add Ask Music, a conversational tool that works across more than 300 million songs. Users can describe what they want to hear, build custom queues, explore an artist’s motivations or ask who played bass on a song. Your Podcast Lineup will appear on the podcast page with a weekly spoken preview of recommended shows. Both are coming soon to YouTube Music and YouTube Premium subscribers worldwide. YouTube will let US users create custom feeds through conversational AI prompts on the web, mobile and TVs. Shorts can be arranged episodically with custom thumbnails and sequential playback. GIF comments will arrive in the mobile app globally this fall, and TV users can post comments by voice. Ask YouTube is adding shopping recommendations and comparison tables and will let users ask about products shown but not mentioned in a video. YouTube extended its Coachella streaming rights through 2030 and will pilot live auto-dubbing for English and Spanish in early 2027. It is also launching live co-hosting and creator matchmaking. YouTube said microdramas surpassed 6.5 billion views in the first half of 2026, with watch time up 50 percent year over year and TV views up 90 percent.

ZeroDrift launched Anchor 3.0, a family of small language models that check messages generated by AI agents before they are sent. The models are designed to enforce financial regulations and internal policies fast enough to examine every outgoing message in real time through ZeroDrift’s Enforcement API. The system intercepts communications and checks them against regulatory requirements and company policies, then can flag, rewrite, block or route problematic messages for human review while recording the decision for audit. ZeroDrift recently introduced Guard for Agents, an API service that inserts checks into agent workflows. The family includes Anchor 3.0 Mini, a 9 billion-parameter mixture-of-experts model with 4 billion active parameters for high-volume traffic; the flagship Anchor 3.0, which applies more than 200 prebuilt rules covering FINRA, the Securities and Exchange Commission and other regulations and can identify and rewrite violating lines; and Anchor 3.0 Max, a 27 billion-parameter model for long documents, attachments and company policies without fine-tuning. The models were post-trained from Google’s Gemma E4B and Alibaba’s Qwen3.8-27B models. ZeroDrift said its flagship detected more than 90 percent of violations in a FINRA-based benchmark, matching the accuracy of OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 while running more than 100 times faster and at less than one 500th the cost. The benchmark used attorney-labeled data from Surge AI, but ZeroDrift published the benchmark itself, so the figures remain company claims. Founder and CEO Kumesh Aroomoogan said frontier models made it easy to build capable agents, but running them inside a regulated business requires every message to follow the rules before anything goes out.

Lookout launched Social Engineering Protection, a mobile module that analyzes texts and calls for signs an employee is being manipulated into handing over credentials or money. Lookout said email security tools and awareness training were not built for conversations on SMS, voice calls and messaging apps, where corporate inspection tools often have no reach, and generative AI has made personalized lures cheap. The first component reads incoming text and multimedia messages on iOS and Android, including Rich Communication Services, flags malicious links and weighs message intent to catch approaches with no link. Voice coverage works on live calls and voicemail, transcribing what a caller says to surface scam patterns and checking the voice for signs it was cloned or machine-generated. The third component assesses phone numbers using mobile identity signals and number attributes to separate legitimate callers from suspicious ones; administrators can block known or suspicious numbers across a workforce from one console. Praveen Mamnani, Lookout’s chief product officer, said frontier AI has changed the economics of cyber offense, making sophisticated social engineering accessible to mainstream attackers, and that existing defenses were built before attackers could run the same approach across text, voice and messaging at once. Social Engineering Protection is a native add-on to Lookout’s Mobile AI Security Platform, and existing customers can enable it from the admin console they already use without deploying an additional agent or infrastructure.