Anthropic's Mythos AI model has demonstrated the ability to autonomously discover thousands of zero-day vulnerabilities in major operating systems and web browsers, including some that remained hidden for over 20 years. Announced on April 7, 2026, Mythos generates working exploit code in hours at low cost. While Anthropic restricts the model to a defensive consortium, similar open-weight models could emerge within 12 to 18 months, collapsing the gap between discovery and weaponization. Security leaders are urged to adopt Zero Trust Architecture to counter threats operating at both human and machine speed.
Anthropic's Project Glasswing consortium uses the Claude Mythos Preview model to find vulnerabilities in critical open-source software, but proprietary software faces even larger risks. AI can now analyze closed-source binaries and reconstruct source code, contributing to an eight-fold increase in edge device zero-day exploits in 2025. Over 40% of exploited vulnerabilities involved end-of-life products that will never receive patches, extending risks to medical devices, industrial controllers, vehicles, and legacy enterprise applications.
Former LSU football coach Brian Kelly is using Anthropic's Claude AI assistant to prepare for his next coaching job interview. Kelly asks Claude questions to build answers for meetings with athletic directors, noting that Claude is better at predicting things outside the lines while ChatGPT is more direct. Kelly was fired from LSU in October 2025 after a 34-14 record over three and a half seasons, and he now faces his first season out of coaching since 1991.
ServiceNow unveiled the Otto AI assistant at its Knowledge conference, unifying Now Assist, Moveworks, and AI Experience across the enterprise. The company also introduced autonomous workforce agents for IT, CRM, employee services, and security and risk. These agents can detect anomalies, perform incident triage, assist with sales qualification, and handle HR and legal tasks. Otto is already available in EmployeeWorks and the AI Control Tower, with rollout to other products planned over the next year.
Google DeepMind CEO Demis Hassabis stated that the cost of running AI models, known as inference, will never become essentially free. He explained that as inference gets cheaper, demand scales to match, with uses like millions of agents working together or single agents thinking in multiple directions. Even if energy costs become zero, the physical creation of chips will still create a bottleneck. Hassabis noted that reasoning models consume more compute, and efficiency will remain a key variable for builders and enterprises.
A Stockton University poll shows that 74 percent of New Jersey voters believe AI will eliminate more jobs than it creates, while only 7 percent think it will create more jobs. 41 percent believe AI will make their lives worse, up from 36 percent in 2024. Additionally, 56 percent of voters would support a ban on AI data centers in their hometown, with 64 percent of Democrats, 46 percent of Republicans, and 54 percent of Independents in favor.
Miami-Dade Public Schools has partnered with the Sheriff's Office to equip almost 900 school buses with AI-powered cameras that detect when cars fail to stop for them. The School Bus Infraction Detection Program was relaunched after being discontinued last year due to concerns about transparency and consistency. Recorded footage is sent to the Sheriff's Office for review, and citations with a $225 civil penalty will be issued starting May 18.
John Deere's Chief Technology Officer Jahmy Hindman shared the company's vision of bringing AI to every acre of farmland. AI enables precision agriculture by analyzing data from sensors to monitor each seed's needs, such as water or nutrient deficiencies. The company uses computer vision and machine learning to create detailed yield maps and provide farmers with actionable insights for planting, fertilizing, pest control, and harvesting. Hindman emphasized that AI-powered farming helps increase yields, reduce waste, and conserve resources to support global food security.
Tacton launched the AI Product Modeling Assistant, which acts as an intelligent co-pilot for CPQ modelers by transforming fragmented product documentation into structured, CPQ-ready models. The assistant ingests inputs like spreadsheets, technical specifications, PDFs, and legacy data to generate a model foundation that is typically 70-80% complete. Modelers then review, refine, and approve outputs before synchronizing into Tacton. This approach can reduce modeling effort by up to 50% and shorten implementation timelines.
Key Takeaways
- Anthropic's Mythos AI autonomously discovers thousands of zero-day vulnerabilities in major OS and browsers, generating exploit code in hours.
- Similar open-weight models to Mythos may appear within 12 to 18 months, collapsing the gap between discovery and weaponization.
- Proprietary software faces larger AI security risks than open source, with an eight-fold increase in edge device zero-day exploits in 2025.
- Over 40% of exploited vulnerabilities involved end-of-life products that will never receive patches.
- Former LSU coach Brian Kelly uses Anthropic's Claude to prepare for his next coaching job interview.
- ServiceNow unveiled Otto AI assistant and autonomous workforce agents for IT, CRM, employee services, and security and risk.
- Google DeepMind CEO Demis Hassabis says AI inference cost will never be free due to chip bottlenecks and scaling demand.
- 74% of New Jersey voters believe AI will eliminate more jobs than it creates; 56% support a ban on AI data centers in their hometown.
- Miami-Dade schools will equip nearly 900 buses with AI cameras to issue $225 citations to drivers who fail to stop for school buses.
- John Deere uses AI for precision agriculture, monitoring each seed's needs to increase yields and conserve resources.
Anthropic's Mythos AI finds thousands of hidden bugs fast
On April 7, 2026, Anthropic announced that its Mythos AI model can autonomously discover thousands of zero-day vulnerabilities in major operating systems and web browsers, some undetected for over 20 years. It generates working exploit code in hours instead of weeks, at low cost. While Anthropic restricts Mythos to a defensive consortium, similar open-weight models may appear within 12 to 18 months, collapsing the traditional gap between vulnerability discovery and weaponization. Security leaders must now design architectures that address threats operating at both human and machine speed, and Zero Trust Architecture is the only model that holds against both external and internal AI threats.
Proprietary software faces bigger AI security risks than open source
Anthropic's Project Glasswing consortium uses the Claude Mythos Preview model to find vulnerabilities in critical open-source software, but this addresses only the most visible part of the problem. Proprietary software, hardware, and protocols carry a far larger accumulation of undiscovered bugs because their security has relied on obscurity. AI can now analyze closed-source binaries, reconstruct source code, and find vulnerabilities, as seen in the eight-fold increase in edge device zero-day exploits in 2025. Over 40% of exploited vulnerabilities involved end-of-life products that will never receive patches, and the risk extends to medical devices, industrial controllers, vehicles, and legacy enterprise applications.
Brian Kelly uses AI assistant Claude to prepare for next coaching job
Former LSU football coach Brian Kelly is using Claude, an AI assistant from Anthropic, to prepare for his next coaching job interview. Kelly asks Claude questions to build answers that will help him when he meets with athletic directors. He says Claude is better at predicting things outside the lines, while ChatGPT is more direct. Kelly was fired from LSU in October 2025 after going 34-14 in three and a half seasons, and he is now facing his first season out of coaching since 1991.
ET Most Innovative AI Products Awards 2026 highlight retail transformation
The ET Most Innovative AI Products Awards 2026 recognizes how AI is transforming retail and consumer experiences through personalization, intelligent operations, and measurable impact. The category honors solutions that redefine how brands engage with consumers across digital and physical touchpoints. It brings together retailers and technology providers to drive scalable, real-world innovation in India's evolving retail landscape.
New Jersey voters support AI ban and data center restrictions
A Stockton University poll shows that 74 percent of New Jersey voters believe artificial intelligence will eliminate more jobs than it creates, while only 7 percent think it will create more jobs. 41 percent believe AI will make their lives worse, up from 36 percent in 2024. Additionally, 56 percent of voters would support a ban on AI data centers in their hometown, with 64 percent of Democrats, 46 percent of Republicans, and 54 percent of Independents in favor.
ServiceNow unveils Otto AI assistant and autonomous workforce agents
At its Knowledge conference, ServiceNow unveiled updates to its workflow management platform, including the Otto AI assistant that unifies Now Assist, Moveworks, and AI Experience across the enterprise. The company also introduced autonomous workforce agents for IT, CRM, employee services, and security and risk. These agents can detect anomalies, perform incident triage, assist with sales qualification, and handle HR and legal tasks. ServiceNow Otto is already available in EmployeeWorks and the AI Control Tower, with rollout to other products planned over the next year.
AI cameras on Miami-Dade school buses to issue traffic tickets
Miami-Dade Public Schools has partnered with the Sheriff's Office to equip almost 900 school buses with AI-powered cameras that detect when cars fail to stop for them. The School Bus Infraction Detection Program was relaunched after being discontinued last year due to concerns about transparency and consistency. Recorded footage is sent to the Sheriff's Office for review, and citations with a $225 civil penalty will be issued starting May 18. Drivers must stop for school buses with flashing red lights on all roads except divided highways with a raised barrier or unpaved median at least 5 feet wide.
ChatGPT users go viral with clumsy scribbly AI image trend
A new trend on ChatGPT has users requesting intentionally basic and terrible-looking doodles, with prompts asking for the most clumsy, scribbly, and utterly pathetic images possible. The trend started when a Korean creative director shared a prompt that went viral after OpenAI CEO Sam Altman reshared it. The resulting images look like crude Microsoft Paint drawings from the 1990s, with cartoonish features and chaotic squiggles. ChatGPT competitor Grok Imagine has also added a Scribbli template for similar rough-hewn images.
John Deere CTO explains how AI transforms farming for every seed
John Deere's Chief Technology Officer Jahmy Hindman shared the company's vision of bringing AI to every acre of farmland in a Pioneers of AI interview. He explained that AI enables precision agriculture by analyzing data from sensors to monitor each seed's needs, such as water or nutrient deficiencies. The company uses computer vision and machine learning to create detailed yield maps and provide farmers with actionable insights for planting, fertilizing, pest control, and harvesting. Hindman emphasized that AI-powered farming helps increase yields, reduce waste, and conserve resources to support global food security.
Google DeepMind CEO says AI inference cost will never be free
Demis Hassabis, CEO of Google DeepMind, said that the cost of running AI models, known as inference, will never become essentially free. He explained that as inference gets cheaper, demand scales to match, with uses like millions of agents working together or single agents thinking in multiple directions. Even if energy costs become zero through fusion or superconductors, the physical creation of chips will still create a bottleneck. Hassabis noted that reasoning models consume more compute, and efficiency will remain a key variable for builders and enterprises.
Tacton launches AI Product Modeling Assistant to speed up CPQ
Tacton launched the AI Product Modeling Assistant, which acts as an intelligent co-pilot for CPQ modelers by transforming fragmented product documentation into structured, CPQ-ready models. The assistant ingests inputs like spreadsheets, technical specifications, PDFs, and legacy data to generate a model foundation that is typically 70-80% complete. Modelers then review, refine, and approve outputs before synchronizing into Tacton. This approach can reduce modeling effort by up to 50% and shorten implementation timelines.
Sources
- Mythos Outside, Agents Inside: The Zero Trust Answer to AI on Both Sides of the Firewall.
- Open Source is the Tip of the Iceberg: Why Proprietary Software, Hardware and Protocols Face Greater AI-Driven Security Risk
- Brian Kelly reveals how AI is helping him prepare for next coaching job
- ET Most Innovative AI Products Awards 2026: Recognising AI-driven transformation in Retail & Consumer Expe
- Stockton Poll: New Jersey Voters Support Artificial Intelligence Ban
- ServiceNow continues its AI transformation with an integrated experience
- AI cameras on school buses to help issue traffic tickets in Miami-Dade
- ChatGPT Trend Has Users Requesting ‘Clumsy, Scribbly And Pathetic’ AI Images
- John Deere CTO on AI's Role in Farming
- Cost Of Inference Will Fall, But It Won’t Go To Zero: Google DeepMind CEO Demis Hassabis
- Tacton Launches AI Product Modeling Assistant to Accelerate CPQ Implementation
Comments
Please log in to post a comment.