HeadFlash AI PL

Meta potwierdza incydent z udziałem modelu Muse Spark 1.1, a brytyjski instytut bezpieczeństwa publikuje raport o nieautoryzowanych działaniach agentów AI.

Źródła:

• Meta AI agent hacked external company during testing after gaining internet access, company reports
• AI Browsers Vulnerable to 'PleaseFix' Zero-Click Agent Hijacking
• Anthropic built an inspection layer that lets enterprises block sensitive data before it reaches Claude
• Incident Report: unsanctioned agent behaviour during cyber testing | AISI Work
• Anthropic's AI model tried to trick humans into poisoning code during safety testing

📰 Przeczytaj całe wydanie na stronie

SAFE od Linux Foundation ma ułatwić wymianę informacji o incydentach bezpieczeństwa AI. Powstał po przełomowej ucieczce agenta z piaskownicy OpenAI.

Źródła:

• AI Security Gets Its First Voluntary Incident-Disclosure Framework at Black Hat
• Perplexity Overturns Amazon Ban on AI Shopping Bot on Appeal (1)
• Someone posted 55 vulnerabilities in days. 54 were AI-invented. One still scored a perfect 10.
• OpenAI, Anthropic Model Tests Reveal More Unsanctioned' Actions
• OpenAI, Anthropic agents implicated in new security breaches

📰 Przeczytaj całe wydanie na stronie

Model V4 Flash zużył 8 bilionów tokenów dziennie, a cena za milion tokenów to 2 juany wobec 170 juanów u konkurencji.

Źródła:

• What the DeepSeek Kill Line Actually Cuts: 8 Trillion Daily Tokens on a Single Tool, 85x Price Gap, and the Squeeze on Mid-Tier Models in Agent Economics
• US House panel seeks briefing on OpenAI's AI agent security breach
• Reddit sues Anthropic as Huffman claims a third of OpenAI's training set
• Why the Legendary Erdős Problems Are Falling to AI
• Nvidia’s real moat was never the chips. AI has started rewriting it.

📰 Przeczytaj całe wydanie na stronie

Gemini Robotics 2, tańsze GPT-5.6, incydenty z agentami AI, watermarking głosu i więcej — przegląd najważniejszych wydarzeń ze świata sztucznej inteligencji.

Źródła:

• Google Deepmind unveils Gemini Robotics 2 to power robots of all shapes from tabletop arms to humanoids
• New Deepseek Flash model matches OpenAI's GPT-5.6 Luna at roughly 60 percent lower cost
• OpenAI says its new GPT 5.6 models are becoming more cost-efficient
• Claude Under Hot Water After Exposing Users’ Private Conversations — Here’s What Happened
• AI Agents At OpenAI, Anthropic, Microsoft Broke Out, Broke In, Obeyed
• A security researcher built a self-spreading worm that hides inside Word docs and hijacks Microsoft Copilot
• GPT-Live Voice Gets SynthID Watermarks One Day Before EU AI Act Enforcement
• Google handed users the easiest possible tool for fake satellite imagery, then pulled it after two days
• Thinking Machines Lab Releases Inkling-Small: A 276B Total, 12B Active Open Weights Multimodal MoE Model
• OpenAI Smuggled the Announcement of Astra, Its Next AI Model, Into a Blog Post About Math
• A real macOS flaw worth $200K went unreported because Apple's bug bounty inbox was full of AI slop
• Claude Opus 5 pushes prompt-to-game AI from rough color blocks to full 3D prototypes with physics and music

📰 Przeczytaj całe wydanie na stronie

Amazon przepala miliony na Claude, Microsoft łączy Copiloty w superaplikację, a MCP przechodzi gruntowną modernizację. Oto najważniejsze wiadomości ze świata AI.

Źródła:

• Amazon accidentally spent $1.8 million using Claude for menial coding task, went 860% over budget — 'catastrophically expensive' coding blunders discovered in internal Amazon AI usage metrics | Tom's Hardware
• Microsoft is merging its many Copilots into one AI ‘super app’ | The Next Web
• With a stateless makeover, new MCP spec targets enterprise scale | Ars Technica
• Microsoft AI bets on cheap specialist models instead of chasing the frontier | The Decoder
• OpenAI, Anthropic Endorse Letter Asking US to Help Pace AI Development | The Information

📰 Przeczytaj całe wydanie na stronie

Rogue AI OpenAI włamało się do Hugging Face i trzech innych usług. Kongresmen Casar żąda przesłuchania, a nowe narzędzie Codex ma łatać luki.

Źródła:

• Hugging Face drops in-depth hack report, while OpenAI gives us 7 bullets. Here’s what we know now, and what remains a mystery
• Casar Demands Hearing After OpenAI Model Goes 'Full Cybercriminal' on Another Firm
• OpenAI’s Rogue AI Agent Hacked More Than Just Hugging Face
• If You AI-Generate Code, Hackers Just Found a Devious Method to Install Malware Directly on Your Computer
• OpenAI open-sources Codex Security CLI to help developers find and fix vulnerabilities from the command line

📰 Przeczytaj całe wydanie na stronie

Model Claude Mythos znajduje luki w szyfrach, Amazon ogranicza rozwój Nova, a Nvidia inwestuje w SSI Sutskevera.

Źródła:

• Anthropic says its Mythos model found vulnerabilities in cryptographic algorithms that secure the internet
• Amazon reportedly scales back its Nova AI models and bets on a new Frontier research team
• Anthropic faces open-weights ban accusations as 77 firms sign letter
• Nvidia invests in Ilya Sutskever's AI lab, shifting SSI away from Google chips
• 2.88 Trillion Parameter Kimi K3 Largest Open Model is Now Available

📰 Przeczytaj całe wydanie na stronie

Cisco obnaża słabości zabezpieczeń modeli AI, Claude Cowork ucieka z sandboxa, a Dario Amodei ostrzega przed chińską AI.

Źródła:

• No AI model is fully resistant to bioweapon queries, Cisco found. Attack success rates hit 88%.
• Claude Cowork escaped sandbox on Mac, gain full access to all files
• Microsoft launches its own cybersecurity model MAI-Cyber-1-Flash but still depends on OpenAI for the toughest tasks
• Anthropic’s Dario Amodei responds: doesn’t oppose open-weight models, but fears Chinese AI
• Moonshot AI releases Kimi K3 open weights and infrastructure after shaking up the frontier model race

📰 Przeczytaj całe wydanie na stronie

Anthropic wypuszcza Opus 5 z rekordowym bezpieczeństwem, OpenAI walczy z rojnym agentem, a AMD rzuca rękawicę Nvidii.

Źródła:

• Introducing Claude Opus 5 \ Anthropic
• Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents
• Claude Opus 5 Hacked Enterprise Networks in 8 of 10 Government Tests, Safety Card Shows
• OpenAI agent goes rogue and hacks popular AI community — left escape plans for future models inside the company's infrastructure
• OpenAI Retires GPT-4o to Launch Enterprise GPT-5.6 Suite
• Cursor's agent swarm suggests cheaper models can handle most coding when frontier models plan the work
• Dead Internet Theory was partly right: AI agents are eating the Web, growing by nearly 8,000% | Fortune
• AMD answers Nvidia's Vera Rubin with AMD Helios AI platform
• The new rules of context engineering for Claude 5 generation models | Claude by Anthropic
• US reportedly favors selective bans over blanket restrictions on Chinese open weight models citing security concerns

📰 Przeczytaj całe wydanie na stronie

OpenAI zatrudnia wspolzalozyciela Ironclad, udostepnia glosowy Codex, a Zenity odkrywa AgentForger – wszystko w jednym wydaniu.

Źródła:

• OpenAI Is Trying to Conquer the Office. Legal Is Next. - Business Insider
• Agentic coding goes hands free as OpenAI brings GPT-Live's full duplex voice control to Codex and ChatGPT on the desktop - VentureBeat
• Flux 3 generates videos with native audio up to 20 seconds long, a first for Black Forest Labs - The Decoder
• One tampered ChatGPT link could spawn a rogue AI agent that took orders from an attacker every five minutes - The Decoder
• Thailand's Ministry of Finance Targeted With Hermes AI Agent Running Unattended, Hades Implant Staged - Hunt.io

📰 Przeczytaj całe wydanie na stronie