Anthropic is investigating unauthorized access to its Mythos AI model, a cybersecurity tool designed for enterprise use. A small group of users, linked to a private Discord channel, gained entry on April 7 by guessing the model's online location using knowledge of Anthropic's previous formats. The breach occurred through a third-party vendor environment, though Anthropic reports no impact on its own systems. UK AI Minister Kanishka Narayan has warned businesses about the potential dangers, noting the model successfully completed 32-step cyber-attack simulations in three out of ten attempts.
Major corporations including Amazon, Apple, JP Morgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley have been granted early access to Mythos as part of Project Glasswing. These companies are using the tool to detect vulnerabilities before bad actors can exploit similar AI models. However, CISA has been excluded from the rollout while the NSA and Commerce Department are already testing it, raising questions about coordination within the Trump administration's AI deployment strategy.
Meanwhile, Google's DeepMind is partnering with top consultancies like Accenture, Bain & Company, BCG, Deloitte, and McKinsey to accelerate enterprise AI adoption. The initiative provides early access to Gemini models and aims to help companies scale solutions, addressing the hurdle where only 25% of firms have successfully implemented AI despite its projected $15.7 trillion economic contribution by 2030. Google is also launching a fund for AI adoption consultants to provide feedback and solve client issues ahead of broader Gemini launches.
In other developments, Macy's built an 'Ask Macy's' AI shopping agent in just four weeks using Gemini Enterprise to improve the online customer experience. The agent serves as a digital concierge with a warm, conversational tone. On the hardware front, DigiFlight is deploying its Phoenix Mixed Reality Flight Training Device for Apache helicopter pilots, while a humanoid robot named Lightning broke the Beijing half-marathon world record before breaking itself during the race.
Key Takeaways
['Anthropic is investigating unauthorized access to its Mythos AI model by a group linked to a private Discord channel.', "The breach occurred on April 7 through a third-party vendor environment after users guessed the model's online location.", 'UK AI Minister Kanishka Narayan warned businesses about the significant cyber threats posed by the Mythos model.', 'Mythos successfully completed 32-step cyber-attack simulations in three out of ten attempts during vetting.', 'Major banks including Amazon, Apple, JP Morgan Chase, and Goldman Sachs are early adopters of Mythos under Project Glasswing.', 'CISA has been excluded from the Mythos rollout while the NSA and Commerce Department are already testing it.', 'Google DeepMind is partnering with Accenture, McKinsey, and Deloitte to drive enterprise AI adoption and scale solutions.', "Macy's developed an 'Ask Macy's' AI shopping agent in four weeks using Gemini Enterprise to enhance customer service.", 'A humanoid robot named Lightning shattered the Beijing half-marathon world record before breaking itself during the event.', 'NTT Data shared 12 strategies for integrating AI into Security Operations Centers to reduce incident effort by up to 70%.']Unauthorized group hacks Anthropic's Mythos AI tool
A small group of unauthorized users reportedly gained access to Anthropic's Mythos AI model, a powerful cybersecurity tool designed for enterprise security. Anthropic confirmed it is investigating the breach, which occurred through a third-party vendor environment after the group guessed the model's online location. The group, part of a private Discord channel, has been using the software since it was announced but claims they are only interested in exploring the technology rather than causing harm. Anthropic states there is no evidence that their own systems have been impacted by this unauthorized activity.
UK AI minister warns of Mythos security risks
UK AI Minister Kanishka Narayan has stated that businesses should be worried about the potential dangers of Anthropic's Mythos AI model. The model was vetted by the UK's AI Security Institute (AISI), which warned it poses a significant cyber threat compared to previous models. Mythos successfully completed a 32-step simulation of a cyber-attack created by AISI, solving the challenge in three out of ten attempts. The tool can identify weaknesses in IT systems without human intervention, a task that normally takes human professionals days to complete.
Major banks test Anthropic's exclusive Mythos AI
Anthropic has released its Mythos AI model to a select group of major companies as part of Project Glasswing to prevent misuse by bad actors. The list of early adopters includes Amazon, Apple, JP Morgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley. Treasury Secretary Scott Bessent convened a meeting of senior American bankers in Washington in April to discuss using the model to detect vulnerabilities. Anthropic stated the tool could be weaponized against corporate security if it falls into the wrong hands.
Mythos AI accessed illicitly on announcement day
Anthropic's Mythos AI model was accessed by a small group of unauthorized users on April 7th, the same day the company announced its limited release. The group, which includes members from a Discord channel focused on unreleased AI models, used knowledge of Anthropic's other model formats to guess the online location of the new tool. Bloomberg reported that the group provided screenshots and a live demonstration of the software to the news outlet. Anthropic confirmed they have no evidence that the unauthorized access is impacting their systems or extending beyond the third-party vendor environment.
Anthropic investigates Mythos AI breach report
Anthropic is investigating a report of unauthorized access to its Mythos AI model, which was released to a limited group in April as part of Project Glasswing. The company confirmed the investigation to CBS News after a spokesperson stated the access came through a third-party vendor environment. So far, Anthropic has not detected any breaches outside of the vendor environment or any compromises to their own systems. The model was shared with major companies like Amazon, Apple, Cisco, JPMorgan Chase, and Nvidia to help them harden their defenses before bad actors can exploit similar AI models.
Unauthorized users breach Anthropic's Mythos AI
A group of unauthorized users reportedly gained access to Anthropic's Claude Mythos Preview AI model despite the vendor's efforts to keep it out of public hands. The group used knowledge of Anthropic's previous model formats to make an educated guess about the online location of the new system. Bloomberg reported that the group provided screenshots and live demonstrations as evidence and began using the tool on the same day it was publicly announced. Members stated their intent was exploratory rather than malicious, though the incident highlights how easily access controls can be bypassed.
Security experts warn of Mythos access risks
Shane Fry, CTO at RunSafe Security, stated that unauthorized users accessed Anthropic's Mythos model by simply changing a model name. He noted that even if the group's intent is only to explore, it shows how easily these systems can be exposed. The incident highlights the need for enterprises to tightly manage third-party access and maintain visibility when adopting AI security tools. Early-access programs can introduce additional exposure if controls, monitoring, and isolation are not consistently enforced.
Google launches fund for AI adoption consultants
Google's AI lab DeepMind will give early access to Gemini models to select firms that will use the AI tools and provide feedback ahead of launch. Google engineers will also work alongside consulting firms to help solve client issues. This initiative aims to accelerate the rollout of advanced AI systems into the business world by leveraging technical expertise and industry knowledge.
DeepMind partners with top consultancies for AI
Google DeepMind announced strategic partnerships with global consultancies including Accenture, Bain & Company, BCG, Deloitte, and McKinsey to drive enterprise AI adoption. The collaboration focuses on three core areas: enabling scaled, industry-specific AI capabilities, providing early access to DeepMind's frontier models like the Gemini family, and engaging directly with customer CEOs and boards. This move aims to address the hurdle where only 25% of companies have successfully scaled AI solutions despite the technology's projected $15.7 trillion economic contribution by 2030.
DigiFlight brings AI to Apache helicopter training
Columbia-based firm DigiFlight is pushing into AI-assisted training with its Phoenix Mixed Reality Flight Training Device designed for the AH-64 Apache helicopter. The simulator recreates the helicopter's dual-seat layout for pilot and co-pilot or gunner, complete with detailed flight controls, instruments, and cockpit geometry. The device aims to mirror the Apache's operational environment down to its tandem cockpit and mission systems to make training look less like rehearsal and more like the real thing.
NTT Data shares 12 AI strategies for SOC scaling
NTT Data cybersecurity leaders shared 12 practical strategies for integrating AI into Security Operations Centers (SOCs) to reduce incident effort by up to 70%. The strategies include starting with a clear vision, defining use cases, choosing the right AI technologies, and integrating AI into existing workflows. NTT Data also emphasizes the importance of training teams on AI tools, establishing data governance frameworks, and fostering a culture of innovation. By following these strategies, organizations can improve overall SOC efficiency in the evolving cybersecurity landscape.
Chinese marathon robot breaks itself in race
A humanoid robot named Lightning shattered the Beijing half-marathon world record this weekend, completing the race in 50 minutes and 26 seconds. Cyber expert Kurt Knutsson warned that such massive leaps in artificial intelligence necessitate stronger guardrails to ensure safety and prevent future human replacement. During the race, the robot fell over and then started break dancing and bashing itself into pieces. The event took place at the Beijing E-Town Half Marathon and Humanoid Robot Half-Marathon on April 19, 2026.
Nashville man arrested for Grok AI child abuse
A Nashville man, Jerry Dubose II, 47, is facing a child sex abuse charge after authorities connected images generated using Grok artificial intelligence to his account. Dubose was arrested on April 20 and booked into the Downtown Detention Center with a $25,000 bond on a charge of sexual exploitation of a minor. Detectives searched his home in Bellevue after multiple tips about his internet activity were reported to the National Center for Missing and Exploited Children. Grok AI, launched by Elon Musk in 2023, pulls information from content on X and across the web.
M&A helps tech companies fill AI capability gaps
Tech companies are increasingly turning to mergers and acquisitions to fill gaps in their artificial intelligence capabilities, according to an analysis by The Information's Deep Research. The trend is evident across deals of all sizes, from large strategic acquisitions to smaller, targeted purchases. Notable examples include SpaceX acquiring smaller firms to bolster AI capabilities in satellite imagery and autonomous systems, and IBM acquiring AI startups to enhance its Watson platform. The analysis also highlights the importance of M&A in bridging the talent gap for AI development.
Macy's builds AI shopping agent in 4 weeks
Macy's built the 'Ask Macy's' AI agent in 4 weeks using Gemini Enterprise for Customer Experience to improve the online shopping experience. The agent serves as a digital shopping concierge to help customers navigate the almost infinite choice available in e-commerce. Macy's set a new target launch date for the agent at the Shoptalk conference on March 24, 2026, and the teams met for daily virtual standup meetings starting on February 9. The integrated team worked to deliver a reliable, feature-complete agent that provides a warm and conversational tone comparable to in-store service.
CISA excluded from Anthropic's Mythos rollout
The Cybersecurity and Infrastructure Security Agency (CISA) has been left out of Anthropic's Mythos Preview rollout while other federal agencies like the NSA and Commerce Department are already testing it. This exclusion raises questions about coordination in the Trump administration's broader push to deploy AI across government. CISA operates under the Department of Homeland Security, while the NSA falls under the Department of Defense, creating complex relationships when it comes to a tool designed to find vulnerabilities. The timing makes the exclusion even stranger given CISA's mission to defend federal networks.
Mississippi College mandates AI training for law students
Students at the Mississippi College School of Law are now required to learn how to use artificial intelligence as part of their legal education. The law school launched a two-day boot camp for first-year students covering the basics of AI, best practices, ethical considerations, and state and federal regulations. Students must pass an exam at the end of the training. Dean John P. Anderson said the goal is to ensure future lawyers understand how to use AI responsibly from the start of their careers.
Sources
- Unauthorized group has gained access to Anthropic's exclusive cyber tool Mythos, report claims
- Anthropic investigates report of rogue access to hack-enabling Mythos AI
- Unauthorised users hack Anthropic's 'too dangerous to release' AI
- Anthropic’s most dangerous AI model just fell into the wrong hands
- Anthropic investigating possible breach of its Mythos AI model
- Unauthorized Users Reportedly Gain Access to Anthropic’s Mythos AI Model
- Anthropic Probes Alleged Unauthorized Access to AI Security Tool Mythos
- Google Launches $750 Million Fund for Consultants to Adopt AI
- DeepMind Taps Big Consultancies for AI Push
- Apache Training Enters AI-Assisted Era With DigiFlight’s ‘Phoenix’ Simulator
- Real Lessons from Scaling SOC Operations With AI -- Security Today
- Chinese marathon robot falls, break dances itself to pieces
- Nashville man used Grok AI to generate child porn, police say
- How Tech M&A Is Filling AI Gaps
- A New Era for Online Shopping: How Macy’s Built the ‘Ask Macy’s’ AI Agent in 4 Weeks With Gemini Enterprise for Customer Experience
- CISA Locked Out of Anthropic's Mythos AI Security Tool
- Mississippi College law school makes AI training mandatory for students
Comments
Please log in to post a comment.