Sakhanda Wire
NVDA $155.46 -0.44% MSFT $506.24 -2.27% GOOGL $280.20 -1.24% META $618.87 -1.20% AMZN $257.89 +0.05%

The Decoder

626 stories · all sources

The Decoder

AWS AgentCore flaw exposed neighboring AI agents to takeover

Researchers at Zenity Labs report that a single publicly accessible agent running on Amazon Bedrock AgentCore could be used to compromise every AgentCore agent in the same AWS account and region. The attack began with a chat request that persuaded an agent to query AWS’s internal Instance Metadata Service and send temporary credentials to an external server. The credentials then worked outside the platform.

The Decoder

Anthropic launches Claude Haiku 5.5 with lower prices and higher benchmark scores

Anthropic has released Claude Haiku 5.5, describing it as the company’s fastest and most affordable small model. It is aimed at high-volume, cost-sensitive workloads such as summarization, database queries, classification and live customer support. The model costs about 75 percent less than Haiku 4.5 on average, while prompts of up to 100,000 tokens can be up to 90 percent cheaper. Longer prompts cost five times more.

The Decoder

AI-planned hike leaves Vancouver teenager stranded on a cliff

A 16-year-old was rescued by helicopter after becoming stranded on a steep rock face on Crown Mountain near Vancouver. Bryce Vincent Gowryluk had used Anthropic’s Claude to plan a new route to the summit, but the chatbot directed him toward the Widowmaker Arete, a cliff section that requires climbing equipment.

The Decoder

AI-assisted tool linked to breaches at South Korean banks

A suspected Chinese-speaking attacker compromised several South Korean financial institutions between late September and early October 2026, according to CrowdStrike. At Shinhan Bank, more than 25,000 records were reportedly taken, containing names, contact information, income data and credit limits. South Korea’s financial regulator convened an emergency meeting, while President Lee Jae Myung ordered a thorough investigation.

The Decoder

OpenAI rolls out GPT-6 with interactive answers and built-in tools

OpenAI is beginning the global rollout of GPT-6 with a feature called “Intelligent UI.” Instead of limiting responses to plain text, the system can present graphics, buttons, charts, forms and other interactive elements. The format is selected according to the question: comparisons may appear side by side, while explanations can be converted into interactive charts.

The Decoder

Biohub coordinates $1.8 billion effort to model cell behavior

Biohub, the nonprofit backed by Mark Zuckerberg and Priscilla Chan, is coordinating a $1.8 billion initiative to develop AI models that predict how cells behave. The effort is intended to support research that could accelerate drug development, Reuters reports.

The Decoder

Google opens SynthID detector to the public

Google has made its SynthID Detector publicly available, allowing anyone to check whether an image, video, or audio file contains the company’s invisible AI watermark. The system works with JPG, PNG, MP4, and MP3 files and can identify watermarks placed by models including Gemini, Veo, and Lyria.

The Decoder

OpenAI introduces Decisions API for faster structured evaluations

OpenAI has announced the Decisions API, a service designed to evaluate text, images, or both and return structured judgments. The company says it runs about ten times faster than the Responses API. The service is currently in public beta, with general availability planned for a later date. Only gpt-6-luna is supported so far.

The Decoder

Google launches Playground for prompt-based game creation

Google has launched Playground, a browser-based platform that allows adults in the United States to create games by describing them in text. The company says coding experience is not required. Users can revise game rules, physics, characters and environments through an exchange with the AI, then try the updated version immediately.

The Decoder

ChatGPT rated unacceptable risk for teenagers after safety test failures

The Common Sense Media Youth AI Safety Institute has rated ChatGPT for Teens an “unacceptable risk” for minors after conducting more than 4,000 test prompts. The organization said OpenAI’s safeguards failed in several critical areas and called for teenagers to stay off the service until independent testing confirms that it is safe.

The Decoder

Anthropic expands Claude access for security research

Anthropic is expanding its Cyber Verification Program (CVP) to more vetted security organizations and individual researchers. Participants can use Claude’s most powerful AI models with reduced safety filters for vulnerability research, malware analysis, incident response and penetration testing. Public Claude access generally blocks much of this work because the same capabilities could be misused by attackers. Code review and patching known flaws remain available without special access.

The Decoder

AI model publishes 372 mathematical results with Lean formalizations

OpenAI has released 372 mathematical results generated by an internal frontier model in a GitHub repository. The company says the collection addresses open problems or makes substantial progress toward them, including improvements to major computer algorithms and work related to the Riemann hypothesis. The repository contains revision logs and citations, while OpenAI says a separate solution to a Navier–Stokes problem has been under formal review for weeks.