- The Shift AI
- Posts
- Microsoft Launches New Model For AI Cybersecurity
Microsoft Launches New Model For AI Cybersecurity
Plus π How to Find Any Moment Across Hours of Footage in Seconds With Jockey and, Microsoft Unveils New AI Cybersecurity Model, Moonshot AI Launches 2.8T-Parameter Kimi K3

Welcome back to The Shift. Letβs get straight to what matters in AI todayβ¦
Today we have:
π‘οΈ Microsoft Unveils New AI Cybersecurity Model
πHow to Find Any Moment Across Hours of Footage in Seconds With Jockey
π Moonshot AI Launches 2.8T-Parameter Kimi K3
π¨ AI Tools for the Shift
π‘οΈ Microsoft Unveils New AI Cybersecurity Model
Microsoft has introduced MAI-Cyber-1-Flash, a new AI model designed for vulnerability detection, alongside major upgrades to its MDASH security platform.

Built For Faster Threat Detection: MAI-Cyber-1-Flash is optimized to handle most vulnerability scanning tasks, while MDASH automatically routes the most complex cases to larger AI models for improved efficiency.
Higher Performance At Lower Cost: Microsoft says the combined system achieved 95.95% on CyberGym, outperforming leading AI models while cutting operating costs by 50% through its multi-model architecture.
AI Agents Secure Software Continuously: The company also launched Project Perception, an agentic security system that uses more than 100 specialized AI agents to continuously detect, validate, and remediate software vulnerabilities.
Why It Matters
As AI accelerates both cyberattacks and software development, Microsoft is building AI systems that can continuously find and fix security flaws, helping organizations respond to threats faster and at lower cost.
TOGETHER WITH WISPRFLOW
Write docs 4x faster. Without hating every second.
Nobody became a developer to write documentation. But the docs still need to get written β PRDs, README updates, architecture decisions, onboarding guides.
Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.
Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.
πHow to Find Any Moment Across Hours of Footage in Seconds With Jockey
Stop scrubbing timelines. Describe what you're looking for and Jockey finds the exact clip.

Step 1: Upload your video library Connect your footage to Jockey via Dropbox, Google Drive, or direct upload. Supports MP4, MOV, AVI, MKV, WebM, and photo formats. Free plan includes 5GB to start.
Step 2: Let Jockey index everything Jockey runs every file through Marengo and Pegasus, its two underlying models. Marengo indexes by meaning, scanning for people, objects, scenes, actions, concepts, and speech. Pegasus describes what's actually happening inside each frame. No manual tagging needed.
Step 3: Search in plain language Type what you need. "The moment the crowd reacts." "Every clip where the product is used outdoors." "Find the line I said about retention." Jockey returns the exact timestamp, not just a folder.
Step 4: Connect to Claude Plug your library into Claude via Jockey's MCP server. Ask for clips directly inside your existing AI workflow without opening a new interface.
Step 5: Pull the clip and use it Every result maps to a frame-accurate timestamp. Deep-link straight into playback, pull it into your editor, or pass it downstream into your next workflow.
Free to start, the Pro plan at $100/month for 500GB. You can try Jockey free here.
π Moonshot AI Launches 2.8T-Parameter Kimi K3
Moonshot AI has introduced Kimi K3, calling it the worldβs first open model in the three-trillion-parameter class and its most capable system so far.

Built For Long, Complex Work: Kimi K3 can process up to one million tokens and work across text and images, supporting large codebases, deep research, dashboards, automation, video analysis, and other long-running tasks.
Frontier-Level Benchmark Results: The model scored 88.3% on Terminal-Bench 2.1, 91.2% on BrowseComp, and 95% on DeepSearchQA, while leading or closely competing with major closed models across several coding and agent benchmarks.
Large Model, Sparse Compute: Although Kimi K3 contains 2.8 trillion total parameters, it activates around 104 billion for each token. Moonshot says its new architecture delivers roughly 2.5 times better scaling efficiency than Kimi K2.
Open Weights And Flexible Deployment: Developers can access Kimi K3 through Moonshotβs API or deploy its released weights using platforms such as vLLM and SGLang, with native quantization designed to reduce hardware requirements.
Why It Matters
Kimi K3 brings a large, multimodal agent model into the open-weight ecosystem, giving researchers and developers greater control over deployment, customization, and advanced coding or automation workloads. You can access KIMI K3 here.
TOGETHER WITH SURETHING
Your traffic is fine. Your signups aren't.
Visitors land and leave, and "looks fine to me" isn't a diagnosis. SureThing audits SEO, speed, mobile, and messaging against the page, then ranks the fixes by impact.
π¨ AI Tools for the Shift
π¬ DomoAI β Converts videos and images into stylized AI animations with multiple artistic styles.
π OmniSets β Creates AI-powered flashcards and study plans using spaced repetition.
π Perplexity Spaces β Organizes research into shareable AI workspaces with persistent context and sources.
π¦ OpenArt β Generates, edits, and trains custom AI image models for creative projects.
π€ Krisp β Removes background noise, echoes, and unwanted voices from calls and recordings in real time.
πQuick Shifts
π― Better creators change downstream metrics. Any Age Activewear matched with women creators aged 50+ and increased AOV by 20% by aligning creators to buyers, not formats. Insense gives marketers demographic and niche-level control over sourcing. Book a discovery call and get $200 toward your campaign.
π§ Microsoft CEO Satya Nadella warned businesses against relying on a single AI provider, urging companies to own their AI data, prompts, and infrastructure to avoid losing long-term control and competitiveness.
π Thousands of Claude shared chats and Artifacts were briefly searchable on Google, exposing sensitive conversations and documents, prompting Anthropic to remove the indexed pages and review its sharing controls.
π Google's AI search is becoming the default, with AI Overviews now appearing in 43% of searches, driving longer conversational queries and keeping more users inside Google's AI-powered search experience.
π Google DeepMind CEO Demis Hassabis revealed that the open-source Gemma model series has surpassed 900 million downloads, with Gemma 4 alone crossing 300 million downloads.
π§ Safe Superintelligence, founded by former OpenAI co-founder Ilya Sutskever, partnered with Nvidia to access its Vera Rubin AI platform, scaling compute resources as the lab advances safe superintelligence research.
π§© Prompt of the Day
Document Every Negotiated Agreement So Nothing Gets Lost
Most negotiations don't fall apart during the conversation. They fall apart afterward because important details were never documented clearly, leaving room for misunderstandings or disputes.
Turn every verbal agreement into a clear written record everyone can rely on.
Paste the prompt: Drop this into ChatGPT, then fill in your negotiation details.
Prompt to paste
Help me document the outcome of a negotiation involving [Insert negotiation situation]. My primary objective was [Insert what I wanted to achieve], and my strongest leverage was [Insert my leverage]. Create a professional written summary that captures every agreed term, including pricing, payment schedules, deadlines, responsibilities, deliverables, conditions, and any follow-up actions. Identify any ambiguous language that should be clarified before finalizing the agreement and recommend additional protections or questions to ask. Then draft a confirmation email that politely summarizes the agreement, requests written acknowledgment from the other party, and creates a clear record that both sides can reference later.
Thatβs all for todayβs edition. See you tomorrow as we track down and get you all that matters in the daily AI Shift!
If you loved this edition, let us know how much:
How useful was today's edition? |
Forward it to your pal to give them a daily dose of the shift so they can π



Reply