Good morning. It’s Thursday, February 26. Today we are covering:
Gemini is getting its first agentic capabilities
Instagram to alert parents if teens search for self-harm content
Perplexity may have built a better OpenClaw
Critical Cisco SD-WAN bug exploited in zero-day attacks since 2023
VW in robotaxi push to challenge Waymo and Tesla in self-driving race
Let’s dive in
Gemini is getting its first agentic capabilities
By Allison Johnson via The Verge
Google Gemini is adding task automation on select Pixel 10 models and the Samsung Galaxy S26, letting it start workflows like booking an Uber or assembling a DoorDash/Grubhub order from a simple prompt.
Gemini runs the app in a virtual window, taps through each step using Gemini 3 reasoning, and lets you watch, pause, or take over, while flagging moments that need input (like choosing options or handling out-of-stock items).
Google says this is part of turning Android into an “intelligence system,” with broader automation expected in the next major release (likely Android 17), starting as a limited early preview in the US and Korea across a small set of supported apps.
𝕏: Tons of new AI-powered @Android features announced @SamsungMobile Unpacked, including Gemini automations, smarter Circle to Search, scam detection and more. - Sundar Pichai (@sundarpichai)
Instagram to alert parents if teens search for self-harm content
By Richard Morris via BBC
Instagram will start alerting parents using its supervision tools if their teen repeatedly searches for suicide or self-harm terms, marking the first time Meta proactively notifies parents about these searches rather than only blocking results and pointing teens to help resources.
The alerts roll out next week to families enrolled in Instagram’s Teen Accounts in the UK, US, Australia, and Canada, with global expansion later; notifications may arrive via email, text, WhatsApp, or in-app, and Meta says it will “err on the side of caution” even if some alerts prove unnecessary.
Child safety groups are split: the Molly Rose Foundation warns the “forced disclosures” could panic parents and backfire, while Papyrus and researchers say the impact will depend on whether Meta pairs alerts with useful, immediate guidance and tackles ongoing concerns that platforms can still recommend harmful content.
The best way to reach new readers is through word of mouth. If you click THIS LINK in your inbox, it’ll create an easy-to-send pre-written email you can just fire off to some friends.
Perplexity may have built a better OpenClaw
By Jason Hiner via The DeepView
Perplexity launched Perplexity Computer, a “general-purpose digital worker” that can create and execute long-running workflows through the same interfaces users do, positioning it as a more versatile, easier-to-use take on the emerging personal AI agent wave led by Claude Code and OpenClaw.
The product coordinates tools, files, personal context, agentic web access, coding, and file creation, orchestrating work across 19 models (a mix of open-source and proprietary), and runs inside a secure development sandbox to contain execution.
It’s live on the web at perplexity.ai/computer for Perplexity Max subscribers first ($200/month), with rollout planned for Pro ($20/month) and Enterprise, and introduces consumer per-token billing (Max includes 10,000 tokens, plus an additional 20,000 launch tokens).
𝕏: Perplexity breaking grounds into the Financial sector. Take notice investors. PS: Whoever made this video deserves a raise. - Heisenberg (@Mr_Derivatives)
Critical Cisco SD-WAN bug exploited in zero-day attacks since 2023
By Lawrence Abrams via BleepingComputer
Cisco disclosed a critical (CVSS 10.0) authentication bypass in Cisco Catalyst SD-WAN—affecting SD-WAN Controller (vSmart) and SD-WAN Manager (vManage) in both on-prem and cloud deployments—that’s been exploited as a zero-day to compromise controllers and add malicious rogue peers.
The flaw breaks SD-WAN peering authentication, letting attackers gain a high-privileged internal account, access NETCONF, and manipulate SD-WAN fabric configuration; Cisco Talos tracks the activity as UAT-8616 and says exploitation dates back to at least 2023.
CISA Emergency Directive 26-03 orders U.S. federal agencies to inventory affected systems, pull forensic artifacts, ensure external log storage, hunt for compromise linked to CVE-2026-20127 and potential root escalation via CVE-2022-20775, and patch by Feb 27, 2026 (5:00 PM ET)—with guidance warning against exposing SD-WAN management interfaces to the internet and noting updates are the only complete fix.
𝕏: New: @CISAgov orders agencies to quickly patch serious Cisco SD-WAN device vulnerabilities, including two that the agency says are being exploited in ways that imminently threaten government networks - Eric Geller (@ericgeller)
VW in robotaxi push to challenge Waymo and Tesla in self-driving race
By Kana Inagaki via Financial Times
Volkswagen is accelerating its robotaxi push through its MOIA unit, planning an autonomous taxi launch in Los Angeles later this year with Uber and aiming to position itself as the “European champion” as Waymo and Tesla set the pace in the US-led self-driving race.
MOIA already has 100 test vehicles operating across Germany, Norway, and the US (including 30 in Hamburg), is in talks with additional European cities for a 2027 rollout, and targets 100,000 vehicles by 2033 using a “package” model that bundles leasing, maintenance, training, and fleet-management software for local transport operators.
VW’s sensor-heavy ID. Buzz robotaxi approach mirrors Waymo’s hardware stack but relies on Mobileye for the autonomy “brain”; the company argues it can still build a scalable commercial-vehicle business (citing €10bn+ annual revenue potential within seven years), even as critics warn legacy automakers risk becoming commodity hardware providers and VW trails Waymo’s operational lead (about 1mn km tested vs 125mn+ autonomous miles for Waymo).
Trending in AI
Burger King will use AI to check if employees say ‘please’ and ‘thank you’
OpenAI Says ChatGPT Refused to Help Chinese Influence Operations
Thanks for reading to the bottom and soaking in our Newslit Daily fueled with highlights for your morning.
I hope you found it interesting and, needless to say, if you have any questions or feedback let me know by hitting reply.
Take care and see you tomorrow!
How was today’s email?
