OpenAI launches Images 2.0 as Anthropic faces security scrutiny

Key Takeaways

• OpenAI's Images 2.0 introduced a 'Thinking' mode that allows the AI to research and synthesize information before generating visuals.
• ChatGPT Images 2.0 significantly improved instruction following, accurately placing objects and rendering specific details like clock times.
• The new OpenAI model supports multilingual text rendering in Chinese, Korean, Bengali, Japanese, Hindi, and other languages.
• DALL-E 3 now supports various aspect ratios, higher resolutions up to 2K, and seamless 360-degree panorama generation.
• Anthropic's new Mythos model has raised major security concerns by uncovering thousands of vulnerabilities in operating systems and browsers.
• Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent held a high-priority meeting regarding the security risks of Mythos.
• Codenotary launched AgentMon and AgentX to provide real-time monitoring and automated vulnerability fixes for AI agents.
• Bitget is transforming its trading bots into AI co-pilots to help retail traders execute strategies via natural language.
• Together AI offers multi-tenant GPU clusters that allow teams to share hardware while maintaining strict data isolation.
• Norwegian Prime Minister Jonas Gahr Store launched a program to address a reading crisis among youth linked to early AI use in schools.

OpenAI researcher shows ChatGPT image tools get smarter

OpenAI researcher Ayaan demonstrated how ChatGPT's image generation capabilities have evolved to handle complex research and content creation. The new version uses a 'Thinking' mode to perform research, analyze information, and synthesize it into detailed images. This allows the AI to execute full tasks that previous models could not complete. Ayaan showed examples where the AI created product advertisements with estimated values and generated educational infographics about Isaac Newton. The demonstration highlights a major leap forward in the AI's ability to combine world knowledge with visual output.

OpenAI researcher explains better instruction following

OpenAI researcher Jianfeng Gu detailed how ChatGPT Images 2.0 dramatically improves the ability to follow precise instructions for image generation. The new model can accurately place objects and render specific details like clock times that older versions missed. Gu demonstrated the AI creating scenes with exact text placement and multiple objects arranged in specific spatial relationships. The model successfully generated images with five distinct objects positioned correctly relative to one another on a white background. This advancement closes the gap between what users intend to create and what the AI actually produces in the final image.

OpenAI researcher shows infographic and PDF tools

OpenAI researcher Yuguang demonstrated how ChatGPT Images 2.0 can create complex infographics and convert PDF documents into visual assets. The AI follows long detailed instructions to produce professional designs with specific styles and color coding. Yuguang showed how users can upload a PDF file to generate a series of slide-style images or a single poster based on the document content. The system successfully extracted information from a URL to create a visually appealing portrait poster with chart diagrams. This workflow allows users to quickly transform lengthy reports into engaging visual summaries for presentations or marketing.

OpenAI programmer tests multilingual text rendering

OpenAI programmer Boyuan demonstrated how ChatGPT Image 2.0 masters rendering text in multiple languages including Chinese, Korean, Bengali, and Japanese. The model created posters for different cities with accurate text in the requested language and style. Boyuan uploaded a 100-page technical paper and asked the AI to translate it into traditional Chinese and render it as a clear image. The AI successfully translated the document and rendered small text within the image format with high fidelity. This capability allows businesses to create high-quality visual content with accurate text for global audiences without localization costs.

DALL-E 3 adds new aspect ratios and 360 views

OpenAI's DALL-E 3 now offers greater control over image dimensions by supporting various aspect ratios and higher resolutions up to 2K. Researcher Dibya highlighted the ability to generate 360-degree panoramas with seamless stitching that matches edges almost imperceptibly. The model understands prompt requests for specific ratios like 3:1 posters and can output images tailored to different platforms. Dibya demonstrated generating a detailed ocean poster and a panoramic view of Yosemite National Park with high quality. These features give users flexibility to create images for social media, presentations, or artistic endeavors with precise dimensions.

OpenAI unveils advanced thinking image generator

OpenAI has unveiled Images 2.0, a significant advancement that allows AI to think and research before generating visuals. The system moves beyond simple creation to engage in a process akin to research and thinking for more nuanced imagery. Images 2.0 can generate multiple distinct images simultaneously and offers enhanced control over aspect ratios and resolutions. The model supports prompts in multiple languages including Japanese, Korean, Chinese, Hindi, and Bengali for global use. It can create detailed fashion illustrations, architectural blueprints, and animated storyboards with consistent characters.

AI image tool shows off versatile creative styles

A showcase of ChatGPT Images 2.0 reveals the AI's advanced visual generation capabilities using a chameleon metaphor for adaptability. The demonstration featured rapid transitions between vastly different scenes like a wrestler, a cowboy, and a DJ to show versatility. The core message is that the AI can adapt its output to match specific needs and contexts provided in a prompt. This adaptability is crucial for the widespread adoption of AI in creative industries for marketing and educational materials. The video highlights that the AI is becoming a dynamic tool capable of visual storytelling across a wide array of themes.

OpenAI model masters text and complex visual prompts

OpenAI's Images 2.0 model demonstrates significant leaps in understanding complex prompts and rendering accurate text within images. The team showcased how the model could transform a simple photo into a series of logos with consistent aesthetics and style. Images 2.0 can generate text that is legible and contextually appropriate for magazine covers and posters. The model's thinking mode allows it to break down intricate requests for coherent and accurate results. It also exhibits impressive multilingual capabilities to create marketing posters in languages like Japanese and Hindi.

New AI model raises security fears for banks

Anthropic has announced a new AI model called Mythos which has raised significant security concerns among government officials and bank CEOs. Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent held a high-priority meeting to discuss the potential threat Mythos poses to the U.S. financial system. The model was shown to uncover thousands of major vulnerabilities in every major operating system and web browser during a preview. Experts warn that Mythos can identify and exploit vulnerabilities faster than companies can repair them. The model is currently available through a controlled initiative called Project Glasswing for selected tech companies.

Cybersecurity experts warn of new AI threat

Anthropic's latest AI model Mythos is sparking fears from cybersecurity experts and the banking sector about software security threats. The model is equipped with sophisticated capabilities designed for defensive cybersecurity tasks but can identify and exploit vulnerabilities faster than companies can repair them. Anthropic rolled out the Mythos Preview through a controlled initiative called Project Glasswing granting access to major tech companies. Experts warn that the model's advanced coding and autonomous capabilities could dramatically accelerate sophisticated cyberattacks in sectors like banking. Bank of Canada Governor Tiff Macklem confirmed discussions about the model's impact on the global financial system.

Codenotary launches new AI security platforms

Codenotary announced the launch of two new platforms called AgentMon and AgentX to address security and automation challenges for AI agents. The company added 37 new enterprise customers over the past six months including Kroger, Rakuten, and Swiss Life. Codenotary now secures an average of 240 compute instances per customer as organizations move toward continuous verification approaches. AgentMon provides real-time monitoring of agent behavior to track interactions and detect sensitive data leakage. AgentX continuously identifies vulnerabilities and applies verified reversible fixes in real time to reduce manual intervention.

Mexico pyramid shooter linked to online hate groups

The gunman who opened fire on tourists at Mexico's Teotihuacan ruins was a Hitler fanatic who left an AI-generated photo of himself with Columbine school shooters. Julio César Jasso killed a Canadian tourist and shot several others before taking his own life at the archaeological site. Authorities found images of Jasso performing the Nazi salute and an AI photo of him posing with the Columbine shooters. He was also a member of a twisted online subculture known as the True Crime Community that glorifies school shootings. Six Americans and others were taken to the hospital after the shooting incident.

Norway's reading crisis blamed on early AI use

Norwegian Prime Minister Jonas Gahr Store launched a program to deal with a reading crisis among young Norwegians following the mass introduction of AI in schools. Around 500,000 Norwegians cannot read a text message or simple instructions after a decade of giving every child their own digital device. The author argues that AI removes the doing required for children to learn skills like reading and math. He compares AI to a souped-up calculator that prevents students from building mental muscle memory. The author suggests education should stay in the pre-AI world to build the ability to think and do things independently.

Ai-CHA expands globally with low-cost franchise model

Ai-CHA is accelerating global expansion through a low-investment, operationally scalable franchise model for soft serve and milk tea. The brand has confirmed franchisees in Africa reaching 350 and the Middle East reaching 80 with targeted openings expected between Q3 and Q4. Kenya and South Africa have emerged as two of the brand's strongest growth markets with average daily sales above USD 1,000. The menu is built around affordable tea beverages, ice cream, coffee, and other high-frequency products for everyday consumption. The brand's halal-oriented positioning serves as a meaningful tailwind in many countries where those standards influence consumer adoption.

Transparency fixes trust issues in AI search

Nearly two out of three American adults have used an AI-powered search tool but trust remains a defining challenge for the next era. Consumers feel trapped in a walled garden that makes it hard to verify what they are reading. Sixty-three percent of users say they double-check AI search results against other trusted sources like news websites. The research shows that nearly three out of four respondents say AI platforms should always show where their information comes from. Visual evidence like photos of a dish or before-and-after shots would also increase trust significantly. Consumers want AI to show the receipts for its answers.

Lion facing lesson for living in the AI age

Author David Gerber argues that modern existence is losing an embodied aspect due to endless screens and reliance on AI. He compares facing a charging lion to the need for uncomfortable experiences in uncertain times to stay fully alive. Gerber describes how facing down a lion forces a flow state where everything else disappears and you are 100% present. He notes that modern life has lost grittiness with manual vehicles making up only 3.7% of registered vehicles in America. The author suggests that AI is making it easier to live on autopilot and that we need to confront ourselves to avoid drifting along.

Bitget turns trading bots into AI co-pilots

Bitget is doubling down on AI-native trading by launching an AI ecosystem co-creation plan to turn its Agent Hub and GetClaw products into co-pilots. The company aims to narrow the gap between pros and retail traders by making AI a key lever for ordinary users. Bitget's Agent Hub combines Model Context Protocol, REST/WebSocket APIs, and new Skills and CLI modules into a single pipeline. GetClaw is an AI trading assistant that can execute strategies from natural-language instructions and manage positions in real time. Bitget plans to host the world's first AI trading hackathon in the second quarter of 2026.

Shared GPUs solve AI compute conflicts

Together AI's multi-tenant GPU clusters offer a path to cost-effective, scalable AI compute without sacrificing team isolation. The model allows multiple teams to share the same hardware while guaranteeing strict isolation for separate data access and billing. One team's workload will not impact another's thanks to hard quotas and scheduling guardrails. The architecture separates infrastructure into a shared foundation and per-tenant resources on top to ensure zero visibility into adjacent tenants. Quota-based allocation prevents any single team from consuming all GPU capacity and ensures predictable planning.

Sources

• OpenAI's Image Generation Evolves with ChatGPT
• OpenAI's Jianfeng Gu on ChatGPT Images 2.0
• OpenAI's Yuguang on ChatGPT Images 2.0 Capabilities
• ChatGPT Image 2.0 Masters Multilingual Text Rendering
• OpenAI's DALL-E 3 Enhances Image Generation with New Aspect Ratios
• OpenAI's "Images 2.0" Promises Advanced AI Image Generation
• ChatGPT Images 2.0: AI's Chameleon Act
• OpenAI's Images 2.0: AI Masters Text and Visuals
• New AI Model Raises Significant Security Concerns
• Anthropic's latest AI model is sparking fears from cybersecurity experts and the banking sector. Here's why.
• Codenotary Accelerates Enterprise Adoption, Expands Into AI Agent Security and Autonomous Remediation
• Mexico pyramid shooter was Hitler fanatic who left AI pic of him posing with Columbine school shooters
• Contributor: Keep artificial intelligence out of American classrooms
• Ai-CHA accelerates global expansion through a low-investment, operationally scalable franchise model
• AI search has a trust problem. Transparency is the fix
• What Facing Down A Lion Can Teach You About The AI Age
• Bitget unveils AI co-creation push to turn trading bots into ‘co-pilots’
• Shared GPUs, Zero Conflict