Building autonomous AI agents that mutate production state requires moving beyond soft system prompts to a robust zero-trust architecture. To secure Google Agent Development Kit (ADK) workflows against prompt injections and malicious execution, developers must implement hardware-backed cryptographic signatures for database writes, kernel-level sandboxing with gVisor for dynamic code, and deterministic semantic gateways for I/O validation. By enforcing these hard security boundaries at the infrastructure level, you can safely deploy multi-tool AI agents without risking unauthorized data manipulation or server compromise.

HeyGen ported their 18B+ parameter Avatar IV video generation model to Google Cloud's Trillium (v6e) TPUs via torchax and XLA, utilizing FSDP and Ulysses sequence parallelism across an eight-chip mesh. To achieve a 1.86x speedup for real-time streaming, the engineering team pipelined exposed all-to-all collectives, aligned sparse attention block sizes to eliminate mask padding, and bypassed softmax serial dependencies using a precomputed Cauchy-Schwarz upper bound. These custom Pallas kernel and compiler optimizations were deployed only after passing rigorous two-tier quality gates to guarantee byte-identical or mathematically equivalent pixel outputs.

Credentio is a newly released, open-source C++ library from Google that allows developers to integrate high-performance, local-first validation of C2PA Content Credentials into their client and server applications. By processing assets entirely locally with a highly optimized memory footprint, the library delivers instant validation verdicts for multi-gigabyte media files without incurring cloud latency, bandwidth costs, or data privacy risks. The library currently features deep manifest parsing alongside configurable trust list integration, and is available now on Google Source with future plans to support full credential generation and embedding.

As AI coding assistants shift the developer's primary role from writing boilerplate to reviewing and maintaining systems, language choice becomes critical for long-term architectural integrity. Go directly addresses this new paradigm by utilizing its strict compiler, integrated toolchain, and uncompromising readability to provide deterministic guardrails that help AI models self-correct and generate highly standardized code. By enforcing ecosystem-wide consistency and strict backward compatibility, the Go platform empowers engineering teams to efficiently verify, optimize, and maintain high-velocity, AI-generated output in production environments.

Deploying secure, real-time Edge AI on Raspberry Pi is now simplified using LiteRT and lightweight Gemma open models. LiteRT optimizes CPU and GPU performance, delivering fast token speeds for models like Gemma4, enabling real-time local reasoning for robotics. Developers can quickly convert, quantize, and run these models using the lightweight LiteRT CLI tool. Support for Hailo AI accelerators is also coming very soon.

Google Cloud has natively integrated TPU support into the vLLM serving engine, allowing developers to elastically scale high-demand embedding pipelines using Google Kubernetes Engine (GKE). To handle massive 15K+ token contexts for models like Qwen3-Embedding-8B, the engineering team implemented TPU-specific optimizations such as hardware-safe tensor alignment, JAX/XLA compilation pre-warming, and a hybrid StepPool architecture for chunked prefill management. These enhancements achieve near-perfect numerical parity with reference GPU baselines, and developers can immediately leverage the open-sourced setup recipes on the AI-Hypercomputer GitHub to build their own high-throughput semantic retrieval applications.

Agent Anomaly Detection is a new, out-of-band oversight layer for the Gemini Enterprise Agent Platform that analyzes OpenTelemetry traces and tool calls to catch behavioral risks without adding runtime latency to live requests. It utilizes a multi-tiered detection pipeline—combining lightweight statistical scanning with deep LLM-based reasoning—to identify logical anomalies and policy violations grounded in the OWASP Agentic Top 10. Developers can triage these automated findings within Security Command Center or leverage the exposed API to programmatically block subsequent tool calls when an agent breaches defined risk thresholds.

Elon Musk wants rival AI companies to test each other’s models before release, creating an independent safety checkpoint for frontier AI.

The post Elon Musk’s AI Safety Plan: Let Rival Companies Test Each Other’s Models appeared first on TechRepublic.

Agility Robotics unveiled Digit 5 with $300 million in orders, new human-safety features and a 2027 launch, but filings reveal key buyer caveats.

The post Agility’s New Digit 5 Robot Has $300M in Orders — But There’s a Catch appeared first on TechRepublic.

Meta launches Meta One globally with plans starting at $2.99, bundling AI features and premium tools across Instagram, Facebook and WhatsApp.

The post Meta One Pricing Explained: What $2.99, $7.99 and $19.99 Get You appeared first on TechRepublic.

Google left one thing out of the announcement around the new Search profiles features announced today, you can now add a Search profile badge to your website. This allows users/searchers to follow you directly from your website.

China is emphasizing technical controls for AI agents as U.S. leaders debate slowing frontier AI, raising new questions for businesses deploying autonomous systems.

The post China’s Answer to AI Safety: More Controls, Not Slower Development appeared first on TechRepublic.

Agent Platform's evaluation service is now generally available, providing developers with a unified engine to measure agent quality consistently across local development experiments and live production traffic. You can evaluate agents using over 20 pre-built metrics, DeepMind-backed adaptive rubrics, or custom code-based and LLM-as-a-judge metrics stored in a centralized, versioned registry. The service integrates directly into existing workflows via the Agent Platform SDK, agents-cli, and ADK, offering built-in user and environment simulators to automate complex multi-turn testing and streamline CI pipelines.

Meta reportedly plans camera-free Luna smart glasses centered on voice AI, potentially easing recording concerns while leaving privacy questions.

The post Meta’s Camera-Free Smart Glasses Could Solve Privacy Problems appeared first on TechRepublic.

Apple’s iPhone 18 Pro accessory lineup includes three MagSafe cases, two carrying straps, and a FineWoven Wallet with Find My.

The post 6 iPhone 18 Pro Accessories to Check Out in 2026 appeared first on TechRepublic.

Project: Drupal coreProject machine name: drupalDate: 2026-September-16Security risk: Moderately critical 13 ∕ 25 AC:Basic/A:User/CI:Some/II:Some/E:Theoretical/TD:DefaultVulnerability: Third-party librariesAffected versions: >=10.5.0 <10.6.17 || >=11.0.0 <11.3.17 || >=11.4.0 <11.4.7Description: 

The Drupal project uses the CKEditor library for WYSIWYG editing. CKEditor has released a security update that impacts Drupal.

Vulnerabilities are possible if Drupal is configured to use CKEditor for WYSIWYG editing. An attacker that can create or edit content (even without access to CKEditor themselves) may be able to exploit this Cross-Site Scripting (XSS) vulnerability to target users with access to the WYSIWYG CKEditor, including site admins with privileged access.

For more information, see CKEditor's security advisory:

• High-severity Cross-site scripting (XSS) in the engine package

Solution: 

Install the latest version:

Drupal 11

• If you use Drupal 11.4.x, update to Drupal 11.4.7.
• If you use Drupal 11.3.x, update to Drupal 11.3.17.
• Drupal 11.2.x and below are end-of-life and do not receive security coverage.

Drupal 10

• If you use Drupal 10.6.x, update to Drupal 10.6.17.
• Drupal 10.5.x and below are end-of-life and do not receive security coverage.

Note that Drupal 8 and Drupal 9 have both reached end-of-life.

Instructions for contributed modules

Site owners should also review their site following the protocol for managing external libraries and plugins, as contributed projects may use additional CKEditor plugins not packaged in Drupal core.

CKEditor has also released another CVE in today's release that does not affect Drupal, but may affect custom plugins or other usecases:

• Low-severity Cross-site scripting (XSS) in the engine package

Reported By: 

• Piotrek Koszuliński (Reinmar)

Fixed By: 

• catch (catch) of the Drupal Security Team
• Lee Rowlands (larowlan) of the Drupal Security Team
• Mohit Aghera (mohit_aghera), provisional member of the Drupal Security Team
• Jess (xjm) of the Drupal Security Team

Coordinated By: 

• Bram Driesen (bramdriesen) of the Drupal Security Team
• catch (catch) of the Drupal Security Team
• Greg Knaddison (greggles) of the Drupal Security Team
• Lee Rowlands (larowlan) of the Drupal Security Team
• Dave Long (longwave) of the Drupal Security Team
• Jess (xjm) of the Drupal Security Team

The $149 AirPods 5 add stem volume control, broader charging support, a Find My case speaker, and longer battery life over the $129 version.

The post AirPods 5: Should You Pay $20 More for the Wireless Charging Case? appeared first on TechRepublic.

People and companies are using AI hardware and software to make life easier and more accessible for everyone. That’s the case with smartARM, a Toronto-based startup developing a vision-first bionic arm prototype that uses open-source software, with Meta AI glasses as an optional add-on, to explore more intuitive prosthetic control for people living with limb differences.

https://about.fb.com/wp-content/uploads/2026/09/META-4872_BTTxSmartARM_Long_youtube_16x9_1002_Subs.mp4

Many traditional prosthetic hands require users to switch manually between grip patterns for different objects, making it cumbersome to move everyday items. Something as seemingly natural as picking up a glass, and then a spoon, becomes a multistep process. smartARM addressed this by developing AI hardware and software that understands the environment, and  automatically selects a suitable grip for the object – so users can intuitively move between everyday tasks.

smartARM’s prosthetic arm “sees” through a camera embedded in its palm. The team is using DINOv2, Meta’s open-source vision model, to help it recognize everyday objects from just a few reference photos, supporting smartARM’s grip-selection software.This allows it to grip and pick up objects based on direct visual feature reading, without needing prior understanding or additional programming. The arm can adapt almost instantly to objects trained on just a few photos – something that could take weeks previously.

smartARM uses Meta AI Glasses and the Meta Wearables Device Access Toolkit to extend and improve object recognition and understanding, providing more data from the egocentric point of view the glasses offer. Meta AI glasses  give the arm additional context to better understand what the user is trying to interact with. It also lets users add new objects, through the device app on their phones, giving everyone in the smartARM community a way to personalize their prosthetic around their everyday needs.

For the people who wear it, the results are life altering: a bionic arm that works from the first try, with no training required.

Long-time smartARM user and former NFL player Shaquem Griffin described what it feels like to use it: “Of all the tech packed into it, scalability and accessibility matter most to me. This is the future of consumer prosthetics.”

“Building a capable hand is only part of the challenge. Making it intuitive to use matters just as much,” said Hamayal Choudhry, smartARM’s Founder and CEO. “By helping the arm recognize everyday objects and select a suitable grip, we’re working toward an experience where people can focus more on what they want to do and less on how to operate their prosthetic.”

The post Canadian Start-up smartARM Uses AI to Create Intuitive Bionic Prosthetics appeared first on Meta Newsroom.

Google has updated the Search profiles with a new article design, a single login to manage multiple Search profiles and only 10,000 followers on social platforms are required.

Google Cloud API Gateway now offers a model routing feature in Public Preview, allowing developers to dynamically route traffic to models like Gemini, Claude, or OpenAI OSS-GPT without hardcoding endpoints or managing open-source proxies. Developers can easily configure these routing rules directly within their OpenAPI 3.x specifications by mapping virtual model names to specific backend targets on a shared host. Once deployed, the Gateway acts as a serverless ingress layer that accepts standard OpenAI-compatible requests, automatically transcodes the payload to the native schema of the target model, and routes the traffic on the fly.