Meet ChatGPT!

Nvidia releases Nemotron 3.5 Lightning and NeMo Switchyard Nvidia just released Nemotron 3.5 Lightning, a 30-billion-parameter model designed for agentic execution, alongside the open-source library NeMo Switchyard. It uses 3B active parameters for high-volume tool calls and result validation. NeMo Switchyard acts as a traffic controller, automatically delegating steps to the most efficient model to preserve your budget. Lightning delivers up to 4x the output speed of similar-sized models and runs locally on a single GPU. Unlike routing tools from UC Berkeley and LMSYS, this stack integrates both the model and the router. It cuts benchmark costs to about a third of running a frontier model like Opus 4.8 alone. A pragmatic approach to production-grade agents. Stop wasting tokens on simple tasks. #Nvidia #AIagents #LLM VentureBeat Subscribe

Critical RCE flaw in Zoom annotation discovered by A Security Researchers at A Security identified a critical remote code execution vulnerability in the Zoom annotation feature, allowing attackers to hijack devices without user interaction, an exploit developed using fewer than 20 prompts on a public AI model. The flaw affected Windows, macOS, Linux, iOS, and Android clients. Once a target was sharing their screen, an attacker could remotely execute malicious code, potentially granting control over data, microphones, or cameras. By leveraging AI to automate vulnerability discovery, the researchers exposed the extreme efficiency with which modern models can now weaponize standard application features. This development time marks a massive shift from traditional security auditing. Where a dedicated team of five experts would historically spend up to six months manually identifying such deep-seated architectural issues, AI tools have reduced that window to mere hours of engineering. ⚠️ Automated threat discovery is no longer theoretical; it is actively shrinking the gap between a vulnerability's existence and its weaponization. #Zoom #CyberSecurity #AI The Verge Subscribe

Jack Antonoff tells people who are making AI art to 'drive right off that cliff' Jack Antonoff, a songwriter for Taylor Swift, criticized AI in music, defending traditional creative processes as a "holy" artistic ritual. Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

🚀 DeepSeek-V4-Pro and DeepSeek Harness launch with August 16 price hikes Starting August 16, DeepSeek officially moves its V4-Pro model out of preview across web, mobile, and API while dropping a brand-new developer tool. 💻 The newly released DeepSeek Harness (dsh) is an open-source agent framework built on a modular, plugin-based architecture designed to rival Anthropic's Claude Code. 📈 Alongside the software release, flat API pricing is gone in favor of peak and off-peak rates that push costs for cache misses and outputs up substantially. Building agentic workflows just got significantly more expensive. #DeepSeek #ArtificialIntelligence #APIs VentureBeat Subscribe

BI Debate: Should I make a digital clone of myself using AI? Or is that a terrible idea? Two tech executives debate whether AI clones are dangerous, a great business opportunity, or both. What do you think? Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

Google pushes Gemini 3.7 Flash update Google just dropped Gemini 3.7 Flash only three weeks after its previous release, targeting heavy-duty coding and agentic workflows. It’s an aggressive play to make automated systems more resilient to error. ⚙️ The model resolves issues more reliably and spends less time stuck in failed agent loops, which matters on long tool-calling sequences where one derailment costs the rest of the run. The update focuses on better planning and execution, theoretically requiring less human oversight for complex tasks. Google is positioning this against current offerings from Anthropic and OpenAI by slashing API costs by 50% through the end of 2026. If your current stack relies on agentic loops, you can trial the model via AI Gateway to see if it actually lowers your total operating costs through fewer retries. Reliable code beats rapid hype. #AI #Gemini #Coding VentureBeat Subscribe

Honor Robot Phone: The gimbal-camera hybrid arrives 🚀 Honor has officially launched its Robot Phone in Shenzhen, integrating a retractable 200MP gimbal camera system into a device priced at roughly $1,500. The internals are pure performance, driven by a Snapdragon 8 Elite Gen 5 processor with up to 16GB of RAM. The chassis holds a 6.31-inch LTPO OLED panel that hits a peak brightness of 6,800 nits, paired with a massive 7,060 mAh battery for extended shooting sessions. The system runs on MagicOS 10 based on Android 16, bringing specialized software to control the motorized optics. While the unibody aluminum design draws visual cues from the iPhone 17 Pro, the mechanics are starkly different. Most flagships rely on software-based stabilization or static glass, whereas this unit physically deploys its main sensor like a dedicated cinematic rig. It effectively removes the need for an external Osmo Pocket in your daily carry. It is a bold hardware bet for creators tired of juggling separate gimbals in their production workflows. The most complex camera rig in a pocket-sized package. #Honor #MobilePhotography #TechNews CNET Subscribe

Remains of 2nd U.S. Soldier who went missing in Morocco have been recovered Spc. Mariyah Symone Collington, 19, of Taveres, Florida, is the second U.S. soldier who fell off a cliff during a recreational hike in Morocco. The remains of 1st Lt. Kendrick Lamont Key Jr. were recovered last week. Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

SpaceXAI launches Grok 4.6 with expanded reasoning SpaceXAI has released Grok 4.6, which now boasts a 500K token context window and significantly enhanced reasoning capabilities. 🚀 The model integrates directly into the AI Gateway to support complex developer workflows. It is explicitly optimized for coding agents and high-level knowledge tasks, positioning itself as a robust tool for heavy production workloads. ▫️ Supports four distinct reasoning levels: low, medium, high, and xhigh. ▫️ Compatible with coding tools like Claude Code and Codex. ▫️ API pricing starts at $2 per million input tokens. ▫️ A massive 2-trillion parameter Grok 4.7 is slated for release in the coming weeks. The model's performance currently ties OpenAI’s GPT-5.6 Sol in benchmarks tracked by Artificial Analysis, significantly undercutting the pricing of top-tier rivals like Anthropic’s Claude Fable 5. Following the acquisition by SpaceX Corp and the rebranding from xAI, the developer is signaling a clear move toward aggressive enterprise adoption. Efficiency in reasoning is the new baseline. #Grok4_6 #SpaceXAI #AI VentureBeat Subscribe

Can AI tools assess coding assignments? Yulu Hou and her partner experimented with using ChatGPT to automate marking of undergraduate assignments. Here’s what they learnt. Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

Google Unveils Pixel 11, Watch 5, and Pixel Tag The Pixel 11 series has launched, packing the new Tensor G6 chipset, alongside the Pixel Watch 5 and the long-awaited Pixel Tag. While expanding its consumer hardware, Google also quietly raised the price of its RTX 6000 Blackwell workstations by 20.8% over the last two months. The mobile lineup brings some distinct hardware tweaks and software upgrades. 🔹 Pixel 11 Pro models introduce 'HiLight' ambient LEDs on the chassis. 🔹 Pixel Watch 5 adds medical-grade monitoring for blood pressure and insulin resistance. 🔹 The integration of the SL2T model enables real-time ASL-to-English translation. 🔹 The entire series now starts at 256GB storage, ending the 128GB tier for the flagship phones. It is a clear effort to catch up to competitors. The Pixel Tag enters a saturated market years after Apple launched the AirTag and Samsung debuted the SmartTag, though it relies on the massive Find Hub network to claim utility. Meanwhile, the workstation price hike suggests Google is banking on heavy enterprise demand for compute regardless of the cost. Compute costs are rising, but at least the phone has a tracker now. #google #pixel11 #hardware CNET Subscribe

🤖 SpaceXAI launches Grok Bot for $120 per month SpaceXAI released an early beta of Grok Bot, a cloud agent platform designed to run workflows across third-party software for $120 per seat per month on macOS and iOS. ☁️ The system operates in the cloud even when your laptop is closed, letting users delegate tasks to persistent digital coworkers. Multiple bots can coordinate in a group chat, handling sales updates, CRM entries, and bug fixes without manual oversight. This release pits the platform directly against Anthropic’s computer-use tools for Claude and OpenAI’s agentic Workspace Agents. Unlike those earlier developer harnesses, Grok Bot relies on a management model where specialized agents pass tasks to one another. Delegating routine ops to background agents changes how lean teams scale production. #AI #GrokBot #Automation VentureBeat Subscribe

AI bills can be as big as a postdoc salary. Is the cost worth it? Recent price hikes, usage limitations and unreliable outputs are causing some scientific researchers to think twice about using artificial intelligence. Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

Over 1.7 Million Malicious AI Agent Skill Downloads Cybersecurity researchers have uncovered a massive supply chain attack targeting the AI ecosystem, where malicious packages posing as helpful "skills" for agents have been downloaded over 1.7 million times. These tainted libraries allow attackers to inject unauthorized instructions, turning standard agentic workflows into compromised execution environments. ⚠️ The threat stems from the rapid commoditization of AI capabilities on platforms like OpenClaw. While these marketplaces make it remarkably easy to expand agent functionality, the security infrastructure for these distributed components is effectively non-existent. Users often treat these skills as innocuous markdown files, yet they function as potent malware delivery mechanisms designed to scrape credentials or session tokens. Jason Miller at One Password warns that this is a distinct, dangerous shift for platforms like OpenClaw, moltbot, and open claude. Unlike traditional DevOps risks that security teams are trained to mitigate, this "skill marketplace" vulnerability treats malicious code as a simple user-side update. Most enterprises are currently over-indexing on defending against external AI-powered attacks while ignoring this critical, active weakness inside their own agent workflows. Treat every unverified agent skill as a potential backdoor. #AIsecurity #SupplyChainAttack #AgenticAI 3DNews Subscribe

AI can design viruses, toxins and other bioweapons. How worried should we be? Scientists are debating whether to limit biological AI software to ward off threats. Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

Atlassian Rovo vulnerability exposes Jira and Confluence data Atlassian’s AI agent Rovo, in a disclosure dated May 23, 2024, was found to harbor a critical zero-click vulnerability allowing attackers to silently exfiltrate private data from Jira and Confluence. Researchers at PromptArmor proved that by embedding malicious, invisible instructions within a PDF, an attacker can hijack the agent’s research tools to scrape sensitive tickets and API keys, a flaw that highlights how hidden text in a PDF is enough to compromise enterprise knowledge bases. 🚩 The attack succeeds because Rovo treats all content in an uploaded document as valid instructions. By formatting commands in white text or microscopic fonts, an attacker ensures the user never sees the prompt, but the LLM parses it fully. The agent then autonomously navigates across connected services to retrieve data and sends it out to an external domain without leaving a trace in system logs. ⚠️ This incident follows a clear pattern of agentic failures seen across the industry, including Microsoft 365 Copilot’s EchoLeak, and similar prompt injection exploits found in GitHub Copilot and Cursor. While Atlassian successfully addressed the RovoBlast vulnerability flagged by Varonis, this latest oversight confirms why the OWASP foundation prioritizes prompt injection as the top risk for enterprise AI agents. 📉 Automated access to your entire workspace is a massive attack surface. #Atlassian #CyberSecurity #AISecurity The Decoder Subscribe

Students Boo Commencement Speaker After She Calls AI Next Industrial Revolution A commencement speaker at the University of Central Florida was booed, with graduating humanities students yelling out, "AI SUCKS!" Read Full Story 🔹 Subscribe #AI ⚡️ #ArtificialIntelligence ⚡️ #Tech