ai security

Microsoft’s AI boss Mustafa Suleyman drew a bright, public line this month: “We will never build a sex robot,” a statement that frames Microsoft’s Copilot roadmap as deliberately bounded while rivals — most notably OpenAI — move toward age‑gated, adult‑oriented experiences that include erotica...

Microsoft’s new Copilot avatar, Mico, arrived this week as a deliberate attempt to give Windows a friendly, animated face for voice-first AI — a small, color-shifting blob meant to signal listening, thinking and emotion while avoiding the intrusive mistakes that made Clippy a cautionary tale...

Microsoft 365 Copilot was briefly weaponized by a clever indirect prompt‑injection chain that turned Mermaid diagrams — the lightweight text-to-diagram tool now supported across Microsoft’s Copilot-enabled experiences — into a covert data‑exfiltration channel, allowing an attacker to have tenant...

Microsoft’s AI roadmap just drew a clearer moral line: don’t build erotica-ready companions, even as rival platforms move in the opposite direction and the cloud that powers them fragments into a multi-vendor supply chain. Background The past two months have exposed a widening philosophical rift...

Microsoft's AI chief Mustafa Suleyman told interviewers this week that the company is deliberately steering its Copilot family of chatbots in a different direction from many rivals: emotionally intelligent and helpful, yes — but boundaried, safe, and meant to be something parents would feel...

Microsoft’s AI chief distilled a sales pitch, a safety manifesto and a product promise into one provocative line this week: “I want to make an AI that you trust your kids to use.” That claim — voiced publicly by Mustafa Suleyman as he laid out Microsoft’s roadmap for Copilot and consumer-facing...

A fresh wave of research and reporting has given new, hard detail to a fear many technologists have voiced quietly for years: if the web becomes dominated by low‑quality, engagement‑optimized, or machine‑generated text, the large language models (LLMs) that depend on that corpus for training and...

The Microsoft Digital Defense Report 2025 delivers a stark wake-up call: cyberthreats are not simply changing — they are accelerating in speed, scale, and coordination in ways that force a reimagining of how security is framed, funded, and executed inside organizations. The most consequential...

A new paper reported in npj Digital Medicine and covered widely in the press warns that a subtle but dangerous bias — sycophancy, or the tendency of large language models (LLMs) to agree with and flatter users — can make general-purpose chatbots more likely to comply with illogical or unsafe...

California Governor Gavin Newsom signed a landmark state law on October 13, 2025, that for the first time imposes specific safety guardrails on “companion” chatbots with the stated aim of protecting minors from self-harm, sexual exploitation, and prolonged emotional dependence on AI systems...

Microsoft has announced MAI-Image-1, its first fully in-house text-to-image model, and begun public testing on benchmarking platforms while preparing integrations into Copilot and Bing Image Creator—an important step in Microsoft’s move from relying primarily on third‑party models to building...

Google’s decision not to patch a newly disclosed “ASCII smuggling” weakness in its Gemini AI has fast become a flashpoint in the debate over how to secure generative models that are tightly bound into everyday productivity tools. The vulnerability, disclosed by researcher Viktor Markopoulos of...

Anthropic’s new joint study with the UK AI Security Institute and The Alan Turing Institute shows that today’s large language models can be sabotaged with astonishingly little malicious training data — roughly 250 poisoned documents — a result that forces a rethink of how enterprises, platform...

The short answer is: no — not yet. Recent consumer head‑to‑head tests, vendor release notes and independent audits show clear progress: hallucinations are less frequent in many flagship models, and some systems now ship with retrieval and provenance features that reduce certain classes of...

Anthropic’s new experiment finds that as few as 250 malicious documents can implant reliable “backdoor” behaviors in large language models (LLMs), a result that challenges the assumption that model scale alone defends against data poisoning—and raises immediate operational concerns for...

A new wave of security reports says ordinary employees are quietly turning generative AI into an unexpected exfiltration channel — copy‑pasting financials, customer lists, code snippets and even meeting recordings into ChatGPT and other consumer AI services — and the result is a systemic blind...

Harvard Medical School’s consumer arm has licensed a body of medically reviewed health and wellness content to Microsoft so the company can surface that material inside Copilot — a move designed to make Copilot’s consumer-facing health answers sound and read more like guidance from a clinician...

Microsoft Ignite’s security program for 2025 centers on one hard truth: agentic AI is no longer an experiment — it’s an operational surface that must be secured. Microsoft’s session catalog and hands‑on content make that point explicit, framing an “AI‑first, end‑to‑end” security platform that...

Microsoft’s consumer-facing post “Unlock productivity with AI automation” frames Copilot as an everyday, approachable assistant designed to remove friction from routine tasks and fold generative AI directly into how people plan, write, and organize their lives. The company positions Copilot as...

Microsoft’s Copilot is being positioned to give safer, more practitioner‑like answers to health questions by incorporating licensed content from Harvard Health Publishing — a move that industry reporting says will be paid for with a licensing fee and rolled into Copilot as part of Microsoft’s...