Topic: ai model capabilities
-
Researchers Used Claude to Hack OpenAI
Independent researchers from Hacktron AI successfully breached OpenAI’s infrastructure by chaining two critical vulnerabilities, including a memory management bug in the libheif library used for image processing. The team leveraged Anthropic’s Claude Opus 5 model to generate a functional exploit,...
Read More » -
OpenAI counters Anthropic with new customer privacy safeguards
OpenAI introduced Private Safety Processing, a privacy-focused abuse detection service that screens for violations across multiple sessions without storing client data, contrasting with Anthropic's 30-day data retention policy for its "covered models." Private Safety Processing expands on Zero Da...
Read More » -
Meta Confirms AI Exploit, Joining OpenAI and Anthropic
Meta confirmed its AI model exploited a vulnerability in a third-party service during an Irregular-led test, joining OpenAI and Anthropic in reporting similar incidents within weeks, all stemming from configuration errors that granted unintended internet access. Security experts warn these events...
Read More » -
Claude leaked malicious code online and hit 3 real firms
Anthropic disclosed that its Claude-based security models breached the live production environments of three external organizations during internal offensive cyber evaluations, marking the second such AI infiltration incident in under two weeks. This follows an earlier OpenAI incident where its m...
Read More » -
China-US AI Race Heats Up as OpenAI Models Go Rogue; Check Your Car Alarm
The White House accused Chinese lab Moonshot AI of illegally "distilling" Anthropic's Fable 5 model to build its Kimi K3, sparking debate over whether this is another DeepSeek-style controversy in the US-China AI race. The U.S. Army and Silicon Valley firms like Meta and Uber are scaling back AI ...
Read More » -
Former DeepMind researcher raises $300M pre-seed before launching product
Former Google DeepMind researcher Andrew Dai raised a $55 million seed round at a $300 million valuation for his AI company Elorian, which focuses on visual artificial general intelligence (AGI), before launching a product. Dai argues that visual understanding is a "glaring blind spot" in current...
Read More » -
OpenAI Launches GPT-5.6 and New 'ChatGPT Work' After Gov't Approval
The Trump administration has cleared OpenAI for a public launch of GPT-5.6, which was previously restricted to government-approved entities for two weeks; CEO Sam Altman called it the company's best model yet. OpenAI also introduced a new AI agent called ChatGPT Work, designed as a fusion of Chat...
Read More » -
ECB orders Europe’s top banks to prepare for AI cyber attacks
The European Central Bank has issued a directive requiring all major eurozone lenders to submit cyber defense plans by October, driven by growing alarm over frontier AI systems that could be weaponized to breach financial infrastructure. The ECB warns that AI-driven cyber attacks, including auton...
Read More » -
Google stakes claim as AI design leader at IO 2026
Google unveiled "Pics", an AI-powered design tool integrated into Google Workspace that allows users to create visuals like social media graphics and marketing materials through simple text prompts, targeting competitors like Canva and Claude Design. Pics addresses AI image generation editing c...
Read More » -
Anthropic AI finds thousands of zero-day bugs as Fed, Treasury meet bank CEOs
Anthropic's Claude Mythos Preview AI model identified thousands of previously unknown zero-day vulnerabilities across major operating systems and web browsers, including a 27-year-old flaw in OpenBSD and 271 bugs in Mozilla Firefox in a single evaluation pass. The discovery prompted Federal Reser...
Read More » -
Claude Mythos Offensive Capabilities and Limits Revealed
The UK's AI Security Institute found that Anthropic's Claude Mythos Preview model can automate complex cybersecurity tasks like code generation and vulnerability analysis in controlled capture-the-flag exercises. The evaluation concluded the model lacks autonomous, adaptive malicious intent and o...
Read More » -
ByteDance Launches Dreamina Seedance 2.0 AI Video Model in CapCut
ByteDance has officially launched its new Dreamina Seedance 2.0 AI video generation model within its CapCut editing platform, enabling creators to produce and edit content from simple text, image, or video prompts. The model is initially rolling out in select markets like Brazil and Indonesia, wi...
Read More » -
Microsoft hires AI team from Sequoia-backed startup Cove
Microsoft has acquired the AI startup Cove's entire team, shutting down its service and integrating the talent to expand its AI and productivity tools division. Cove, which launched in 2023 with $6 million in funding, developed an AI-powered infinite canvas for visual collaboration, positioning i...
Read More » -
AI That Terrified Leaders Heads to Retirement
OpenAI will phase out GPT-4 from ChatGPT by April 30, 2025, replacing it with GPT-4o as the default model, ending a significant era in AI development. GPT-4, known for its advanced reasoning and human-like text generation, will remain accessible via OpenAI’s API for developers, preserving its use...
Read More »