Anthropic has released Mythos, an advanced AI model designed to identify thousands of software vulnerabilities in operating systems and browsers. Access is strictly limited to major tech firms and over 40 organizations maintaining critical infrastructure. The White House plans to provide a version to federal agencies, while banks discuss regulations with European authorities. Experts warn this tool could be used for serious cyberattacks if misused.
Mythos can autonomously chain vulnerabilities to create complex exploits, running systematic attack campaigns at a scale no human team could achieve. This dual nature makes it a powerful defensive tool but also a potential weapon. The company is limiting access to prevent misuse, though faster AI iteration could outpace security teams.
In the commercial sector, Adobe introduced the B2B Edition of its Journey Optimizer. This next-gen marketing automation tool uses agentic decisioning to coordinate interactions across email, web, and mobile channels in real time. It allows teams to leverage existing data and trigger adaptive journeys based on customer intent, aiming to improve personalization without requiring a complete system rebuild.
A field experiment in Kenya revealed that AI adoption impacts businesses differently. Small business owners using AI saw revenue increases of 15%, while struggling businesses experienced a 10% decline. The difference depended on whether owners had the judgment to filter misleading AI advice. The study tested a WhatsApp contact connecting users to a customized version of OpenAI's GPT-4.
Snowflake's 2026 report highlights that agentic AI is moving from experiments to production use in sectors like retail and manufacturing. Leaders are realizing that data quality remains the biggest bottleneck for trusted AI systems. Companies are building knowledge bases from archived data and unifying siloed information to enable autonomous agents. The report notes that 58% of companies are already deploying AI while 42% are still planning.
Microsoft has doubled its AI training commitment for New Zealanders, pledging to train an additional 200,000 people by the end of 2028. This builds on a previous 100,000 target, focusing on educators, community leaders, and business workers. Economic modeling suggests AI could add between NZ$76 billion and NZ$108 billion annually by 2038. Microsoft is scaling its Elevate skilling programs to reach these groups through employer-facing training and community outreach.
Z.AI has restricted non-coding use on its developer plans. Users who violate rules three times face permanent account bans, while repeated violations trigger temporary throttling. The move follows similar restrictions from other providers to ensure subscriptions generate value for model training. OpenClaw creator Peter Steinberger noted that cheap access was never intended for general tasks like browsing or email management.
Education faces challenges as schools struggle to teach media literacy in the AI age. A survey found 61% of elementary educators say students struggle to distinguish AI-generated material from real content. Experts warn that AI can create misleading information that harms social cohesion and mental health. Researchers recommend teaching both technical skills and critical thinking about how AI impacts society.
BBVA has set a corporate AI adoption benchmark for companies. Harvard Business Review recognized the bank for its successful strategy of integrating generative AI tools across the institution. The bank trained 250 senior leaders and empowered employees to innovate through decentralized initiatives. Two years after launching the program, over half of employees use AI tools weekly. Success came from facilitating access and encouraging participation rather than imposing strict top-down guidelines.
Cybersecurity threats are accelerating as AI compresses attack cycles from days to minutes. A CrowdStrike report found that threat actors now use AI to analyze reconnaissance data and recommend specific attack vectors instantly. This democratization allows less skilled attackers to conduct sophisticated operations that previously required nation-state resources. Defenders must rely on skilled analysts to interpret AI security tools effectively.
GreenDa AI will join the 2026 World IT Show in Seoul from April 22 to 24. The company will demonstrate its LINDA platform, which uses AI to find overseas buyers and send personalized emails in over 17 languages. More than 800 companies currently use the service to automate their export sales process. The exhibition features 460 companies from 17 countries across 1,400 booths.
Key Takeaways
['Anthropic released Mythos, an AI model limited to major tech firms and over 40 organizations to find software vulnerabilities.', 'Mythos can autonomously chain vulnerabilities to create complex exploits, posing both defensive and offensive risks.', 'Adobe launched the B2B Edition of Journey Optimizer to automate customer journeys using agentic decisioning across multiple channels.', 'A Kenya study showed AI increased revenue by 15% for some small businesses but decreased it by 10% for struggling ones.', "Snowflake's 2026 report indicates 58% of companies are deploying AI, with data quality remaining the biggest bottleneck.", 'Microsoft pledged to train an additional 200,000 people in New Zealand by 2028, aiming for NZ$76-108 billion in economic impact by 2038.', 'Z.AI banned non-coding usage on its developer plans, imposing permanent bans after three rule violations.', '61% of elementary educators report students struggle to distinguish AI-generated content from real material.', 'BBVA achieved high AI adoption with over half of employees using AI tools weekly through decentralized initiatives.', 'AI has reduced cyberattack cycles from days to minutes, allowing less skilled attackers to conduct sophisticated operations.']Anthropic launches Mythos AI with strict access limits
Anthropic released its advanced AI model Mythos through a controlled program called Project Glasswing. The model can find thousands of vulnerabilities in operating systems and browsers, which experts warn could be used for serious cyberattacks. Access is limited to major tech firms and over 40 organizations that maintain critical software infrastructure. The White House plans to make a version available to federal agencies, while banks discuss regulations with European authorities.
Mythos AI poses dual threat as defensive and offensive tool
Anthropic's Mythos model can autonomously discover and chain software vulnerabilities to create complex exploits. This capability makes it a powerful defensive tool but also a potential weapon if misused by attackers. The company is limiting access to prevent misuse while acknowledging that faster AI iteration could outpace security teams. Mythos can run systematic attack campaigns at a scale no human team could achieve.
Adobe introduces next-gen B2B marketing automation
Adobe launched the B2B Edition of its Journey Optimizer to help marketing teams automate customer journeys. The new tool uses agentic decisioning to coordinate interactions across email, web, and mobile channels in real time. It allows teams to leverage existing data and trigger adaptive journeys based on customer intent. This aims to improve personalization and accelerate sales qualification without requiring a complete rebuild of current systems.
AI boosts profits for some businesses but hurts others
A field experiment in Kenya showed that small business owners using AI saw revenue increases of 15% but struggling businesses saw a 10% decline. The difference depended on whether owners had the judgment to filter misleading AI advice. Leaders must design careful rollouts to avoid widening performance gaps between successful and struggling entrepreneurs. The study tested a WhatsApp contact connecting users to a customized version of OpenAI's GPT-4.
Three data trends shape AI adoption across industries in 2026
Snowflake's 2026 report highlights that agentic AI is moving from experiments to production use in sectors like retail and manufacturing. Leaders are realizing that data quality remains the biggest bottleneck for trusted AI systems. Companies are building knowledge bases from archived data and unifying siloed information to enable autonomous agents. The report notes that 58% of companies are already deploying AI while 42% are still planning.
Microsoft doubles AI training commitment for New Zealanders
Microsoft pledged to train an additional 200,000 people in New Zealand by the end of 2028 on top of a previous 100,000 target. The program targets educators, community leaders, and business workers to build an AI-ready workforce. Economic modeling suggests AI could add between NZ$76 billion and NZ$108 billion annually by 2038. Microsoft is scaling its Elevate skilling programs to reach these groups through employer-facing training and community outreach.
Z.AI restricts non-coding use on its developer plans
Z.AI updated its policy to ban non-coding usage on its subsidized coding plan subscriptions. Users who violate rules three times face permanent account bans, while repeated violations trigger temporary throttling. The move follows similar restrictions from other providers to ensure subscriptions generate value for model training. OpenClaw creator Peter Steinberger noted that cheap access was never intended for general tasks like browsing or email management.
Schools struggle to teach media literacy in the AI age
Rising AI use is forcing schools to update media literacy lessons to help students spot fake content. A survey found 61% of elementary educators say students struggle to distinguish AI-generated material from real content. Experts warn that AI can create misleading information that harms social cohesion and mental health. Researchers recommend teaching both technical skills and critical thinking about how AI impacts society.
BBVA sets corporate AI adoption benchmark for companies
Harvard Business Review recognized BBVA for its successful strategy of integrating generative AI tools across the bank. The bank trained 250 senior leaders and empowered employees to innovate through decentralized initiatives. Two years after launching the program, over half of employees use AI tools weekly. Success came from facilitating access and encouraging participation rather than imposing strict top-down guidelines.
AI compresses cyberattack cycles from days to minutes
A CrowdStrike report found that AI has reduced the time needed for cyberattacks from days to hours or minutes. Threat actors now use AI to analyze reconnaissance data and recommend specific attack vectors instantly. This democratization allows less skilled attackers to conduct sophisticated operations that previously required nation-state resources. Defenders must rely on skilled analysts to interpret AI security tools effectively.
GreenDa AI joins World IT Show to showcase sales automation
GreenDa AI will participate in the 2026 World IT Show in Seoul from April 22 to 24. The company will demonstrate its LINDA platform, which uses AI to find overseas buyers and send personalized emails in over 17 languages. More than 800 companies currently use the service to automate their export sales process. The exhibition features 460 companies from 17 countries across 1,400 booths.
Sources
- Explainer: What do we know about Anthropic's Mythos amid rising concerns?
- Mythos: An AI tool too powerful for public release
- The next frontier in B2B marketing automation with AI
- How AI Helps the Best and Hurts the Rest | MIT Sloan Management Review
- 3 Data Trends Shaping the Race to AI Across Industries in 2026
- Microsoft Doubles AI Training Commitment for New Zealanders
- After Claude & Google, Z.AI Restricts OpenClaw-like Non-Coding Usage On Its Coding Plans, OpenClaw Creator Responds
- Schools Play Game of Media Literacy Catch-Up as AI Use Rises
- Harvard Business Review Recognizes BBVA as a Benchmark for Corporate AI Adoption
- When the Attack Comes Faster: What Security Leaders Need to Know About AI-Powered Threats
- GreenDa AI, an AI overseas sales automation startup, to participate in the 2026 World IT Show
Comments
Please log in to post a comment.