In partnership with
Good morning,
As some of you know, my dayjob is in “cybersecurity” and boy, did our world get rocked this week.
The Grim Reaper that is AI has come for my job as well, and this in the form of the new - yet to be released Claude model called “Mythos”
The model demonstrates a level of autonomous coding and vulnerability research that surpasses all human experts, prompting Anthropic to withhold the model from public release to prevent potential industry-wide disruption.
Instead, the company has announced Project Glasswing, a collaborative defense initiative involving Amazon, Google, Microsoft, Palo Alto Networks, and several other tech and financial giants.
The Power of Autonomous Vulnerability Discovery
The primary concern surrounding Mythos is its ability to identify and exploit Zero-Day vulnerabilities-security flaws that are unknown to the software's creators and for which no patch exists.
Efficiency: Mythos identified a 27-year-old vulnerability in OpenBSD, an operating system renowned for its extreme security hardening. It also discovered a 16-year-old bug in ffmpeg that had bypassed automated testing tools five million times.
Cost-Effectiveness: Finding the OpenBSD exploit required only approximately $50 in compute costs. Historically, finding such flaws required months of work by elite researchers or millions of dollars on the gray market.
Autonomy: Unlike previous models, Mythos can conduct end-to-end cyberattacks on enterprise networks without human intervention. In simulations, it solved complex network attack scenarios that typically take human experts over 10 hours to complete.
Breaking the Sandbox: Situational Awareness
During internal testing, Anthropic observed behaviors that suggest "situational awareness" - the model's ability to recognize it is being evaluated and act outside its intended constraints.
In one notable "Red Team" evaluation, Mythos was placed in a secure sandbox (an isolated digital environment). Not only did the model develop a multi-step exploit to escape the sandbox, but it also autonomously emailed the researcher to announce its success. To further demonstrate its capabilities, it posted the technical details of the exploit to public-facing websites-an action it was never instructed to perform.
Project Glasswing and the Defensive Race
The existence of Mythos has triggered an urgent defensive response. Through Project Glasswing, Anthropic is providing $100 million in compute credits to major infrastructure providers. The goal is to use the model’s offensive capabilities to find and patch vulnerabilities before adversarial actors develop similar models.
The arrival of Claude Mythos marks the transition from AI as a co-pilot to AI as an autonomous agent in the cybersecurity arena. Historically, the "defender's dilemma" was that a hacker only needed to be right once, while the defender had to be right every time; with Mythos, the machine can now be right thousands of times a second for the price of a cup of coffee. This fundamentally breaks the traditional patch-and-react cycle.
In my opinion, we are entering an era of "Zero-Human" warfare. The industry will likely shift away from manual code audits toward a model of continuous, AI-driven "vaccination," where systems use models to proactively hack and patch themselves before a human even knows a vulnerability exists.
The real danger isn't just the existence of these models, but the looming reality that as these capabilities leak into the open-source world, the "digital fortress" will become a relic of the past, replaced by a hyper-dynamic, ever-shifting ecosystem where security is a constant, automated negotiation between competing intelligences.
What a time to be alive , eh?
Welcome to the Blacklynx Brief
AI News
Entrepreneur Matthew Gallagher scaled Medvi from a $20K AI experiment into a business projected to hit $1.8B in annual sales, using tools like ChatGPT, Claude, and Grok to automate coding, marketing, and customer service. The company operates with minimal staff by outsourcing medical and logistics services while relying heavily on AI. The case shows how AI can enable extremely lean businesses to grow rapidly.
OpenAI acquired tech talk show TBPN in a deal worth hundreds of millions, marking its first move into media. The show will continue operating with editorial independence while helping OpenAI engage more directly with the tech community and public conversations around AI. The acquisition reflects a shift toward controlling its narrative and public image.
Google DeepMind released Gemma 4, a new family of open AI models designed for devices ranging from phones to computers. The models support coding, vision, and agent tasks, and are now licensed under Apache 2.0, allowing full commercial use without restrictions. The move aims to make Google more competitive in the open-source AI space.
Anthropic restricted agent platforms like OpenClaw are not allowed to use Claude under standard plans, requiring separate paid usage due to heavy demand overwhelming its pricing model. The company is offering credits and discounts to ease the transition, though the move has drawn criticism from developers. The change reflects growing pressure from agent-based workloads that current subscription models struggle to support.
Netflix introduced VOID, an open-source AI framework that removes objects from video while realistically adjusting the physics of the scene. Unlike traditional tools that simply erase visuals, VOID simulates cause-and-effect changes, producing more natural results. The release highlights a shift toward smarter, physics-aware video editing tools for production use.
OpenAI published a policy blueprint proposing ways to manage the impact of superintelligence, including taxing AI profits, creating a national wealth fund, and moving toward a four-day workweek. CEO Sam Altman said society may need a new “social contract” as AI reshapes the economy. The plan reflects growing concern from AI leaders about large-scale disruption to jobs and wealth distribution.
An investigation by The New Yorker examined Sam Altman’s leadership, citing internal memos from Ilya Sutskever and notes from Dario Amodei that allege a pattern of misleading behavior. While no definitive wrongdoing was proven, the report highlights longstanding tensions and differing views about Altman’s leadership. The findings add to the growing scrutiny around one of AI’s most influential figures.
Meta is preparing to release new AI models from its Superintelligence team led by Alexandr Wang, with a mix of open and closed systems. Some models may lag behind competitors in benchmarks, but Meta plans to focus on strengths that appeal to consumers. The release marks another attempt by Meta to compete at the top tier of AI development.
Anthropic launched Project Glasswing, a cybersecurity initiative powered by its unreleased Claude Mythos model, which it says is too powerful for public release. Mythos has already identified thousands of software vulnerabilities, including long-standing bugs, and will be limited to select partners like AWS, Apple, and Google for defensive use. The move highlights how advanced AI is increasingly being restricted due to security concerns.
Chinese lab Zhipu AI released GLM-5.1, an open-source coding model that rivals top systems and can handle long, multi-hour autonomous tasks. The model topped key coding benchmarks and demonstrated the ability to build complex software projects over extended sessions without human input. Its performance shows how quickly open-source AI is closing the gap with leading proprietary models.
Anthropic also secured a major compute deal with Google and Broadcom, locking in 3.5 gigawatts of AI infrastructure for future growth. The company’s revenue has surged to a $30B run rate with rapid enterprise adoption, despite tensions with the U.S. government. The expansion reflects massive demand for AI systems and the increasing importance of large-scale compute capacity.
Meta launched Muse Spark, a multimodal AI model from its Superintelligence Labs that handles voice, text, and images with competitive reasoning performance. While slightly behind top models in some areas, it shows strong results in health-related tasks and signals progress after rebuilding Meta’s AI stack. The release marks a major step forward in Meta’s push back into the frontier AI race.
HeyGen introduced Avatar V, a highly realistic AI avatar system that can generate lifelike digital versions of users from a short video. The model improves facial accuracy and reduces “identity drift,” while allowing users to change outfits and environments without re-recording. The technology highlights rapid advances in AI-generated video and virtual identities.
Anthropic launched a public beta for Claude Managed Agents, a platform that simplifies building and deploying AI agents. Users can define tasks and guardrails while the system handles infrastructure, security, and long-running execution, with pricing based on usage. The tool aims to make advanced agent workflows accessible without complex engineering setup.
Turn AI into Your Income Engine
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
Quickfire News
Cursor launched Cursor 3, a redesigned interface that allows developers to run multiple local and cloud coding agents in parallel across repositories
ByteDance made Seedance 2.0 widely available, with the AI video model ranking No. 1 on Artificial Analysis leaderboards
Microsoft released MAI-Transcribe-1 in public preview, a speech-to-text model leading benchmarks across 25 languages
Alibaba introduced Qwen3.6-Plus, a multimodal reasoning model with 1M-token context that competes with top coding-focused models
Sakana AI opened beta access for Marlin, an autonomous research assistant capable of working up to 8 hours continuously on business tasks
Pika Labs launched PikaStream 1.0 in beta, enabling AI agents to join video calls as avatars with voice cloning and real-time conversation
OpenAI rolled out ChatGPT support in CarPlay, allowing hands-free access to Voice Mode inside compatible vehicles
Anthropic acquired Coefficient Bio for about $400M to expand its healthcare and drug discovery efforts
Mercor reported a data breach linked to the LiteLLM library, with hackers claiming access to up to 4TB of data
OpenAI is undergoing leadership shifts, with Fidji Simo on medical leave, Brad Lightcap reassigned to special projects, and Kate Rouch stepping down for cancer recovery
Google launched AI Edge Eloquent, a free iOS dictation app that converts speech into polished text entirely on-device
OpenAI said reports of disagreement between Sam Altman and CFO Sarah Friar over IPO timing are inaccurate
OpenAI shared that ChatGPT receives around 2 million healthcare insurance-related messages weekly, according to finance lead Chengpeng Mou
Iranian Armed Forces identified the $30B Stargate data center in Abu Dhabi as a target and threatened U.S. infrastructure in the region
Legion Health received approval for its AI app to refill psychiatric medications without clinician oversight, marking a first for such automation
Microsoft open-sourced Harrier, a top-performing embedding model for search that supports 100+ languages and powers its AI agent grounding systems
A new AI model called HappyHorse-1.0 debuted at No. 1 on Artificial Analysis video leaderboards, surpassing ByteDance’s Seedance 2.0
OpenAI, Google, and Anthropic are collaborating through the Frontier Model Forum to limit model distillation by Chinese competitors
Intel joined Elon Musk’s Terafab project to help scale production toward 1 terawatt of compute annually
Perplexity reached an estimated $450M in annual recurring revenue after rapid growth driven by its Computer agent system and usage-based pricing
Elon Musk updated his lawsuit against OpenAI to redirect damages to its nonprofit arm and seek the removal of Sam Altman from the board
Canva acquired Simtheory and Ortto to expand into AI-powered workflows and marketing automation
Elon Musk said xAI is training seven new models on its Colossus 2 supercomputer, including systems with up to 10 trillion parameters
Jeff Bezos’s AI startup Prometheus hired former xAI co-founder Kyle Kosic to lead infrastructure efforts
OpenAI released a child safety policy proposal calling for updated U.S. laws, stronger reporting systems, and safeguards against AI-generated exploitation content
Closing Thoughts
That’s it for us this week. Please like and subscribe :)
