Skip to content
IFAI

AI safety & security

Industry tag1,497 stories

  1. ResearchReach 31

    Meta and Inria's Stable Signature bakes invisible watermarks into AI models

    Meta FAIR and Inria just open-sourced Stable Signature, a new watermarking method for open-source AI image generators. Instead of slapping watermarks on exported images, it roots an invisible signature directly inside the model's decoder, making it survive cropping, compression, and ControlNet fine-tuning. Why it matters: it virtually eliminates false positives while tracing deepfakes straight back to the model that made them.

  2. Generative mediaReach 36

    Meta's AI stickers go completely off the rails

    Meta's new Emu-powered sticker generator is already dodging its own guardrails. While direct sensitive prompts get blocked, simple synonyms and creative phrasing easily churn out kids with weapons, cartoon nudity, and graphic political caricatures. It is a stark reminder of how porous AI content moderation still is when rolled out to billions of users.

  3. IndustryReach 44

    How AI underwriting could fix the insurance industry

    AI is coming for traditional insurance underwriting, tackling painful paper-pushing, pricing miscalculations, and human errors. Machine learning models can analyze unstructured data like text, speech, and images to sharpen risk assessments and flag fraud early. Instead of replacing human underwriters, automated workflows handle low-complexity grunt work so staff can tackle complex cases and retain clients.

  4. ResearchReach 33

    Researchers put AI watermarks to the test—and broke every single one

    University of Maryland researchers just revealed that every existing AI image watermarking method can be cracked. Attackers can easily wipe invisible watermarks clean or even inject them into human-made photos to trigger false positives. While tech giants are betting big on watermarking to fight deepfakes, it's far from a silver bullet.

  5. IndustryReach 43

    Google gives publishers an AI training opt-out without killing search rank

    Google just launched Google-Extended, a tool that lets publishers block their content from training AI models like Bard and Vertex AI while staying visible in Google Search. By adding a simple tag to robots.txt, site owners no longer have to choose between keeping their search traffic or watching their work get scraped for AI models.

  6. AI toolsReach 33

    Google opens generative AI search to teens with extra guardrails

    Google just expanded its generative AI search experience (SGE) to U.S. teens aged 13 to 17 through Search Labs. The rollout includes stronger protections against bullying and illegal substances, alongside an "About this result" feature to explain AI answers. Google is also training the model to stop validating queries with false premises.

  7. IndustryReach 39

    Meta gives its 28 new AI characters their own social profiles

    Meta just launched dedicated Facebook and Instagram profiles for 28 unique AI characters, using famous public figures to give them recognizable faces. U.S. beta users can follow these accounts and chat with them directly through Messenger or Instagram DMs, with all AI-made images clearly watermarked.

  8. IndustryReach 44

    Hollywood writers end strike with historic AI guardrails

    The Writers Guild of America just voted to end its five-month strike after securing a contract that puts strict limits on generative AI. Studios can no longer force writers to use tools like ChatGPT, and AI cannot write or rewrite literary source material. Why it matters: Hollywood just set the first major labor precedent protecting creative jobs from AI automation.

  9. IndustryReach 31

    Google Search caught indexing private Bard chats

    Google Search was just caught publicly indexing shared Bard conversation links, exposing users' chat logs on search result pages. An SEO consultant spotted the blunder, prompting Google to confirm the indexing wasn't intentional and that a fix is actively rolling out to block them.

  10. IndustryReach 61

    Hollywood writers strike ends with strict new guardrails on AI

    Hollywood writers just ended their 148-day strike after reaching a tentative deal with studios that puts strict limits on generative AI. Studios can no longer use AI to write or rewrite scripts, require writers to touch tools like ChatGPT, or treat AI output as source material. Why it matters: It sets the entertainment industry's first binding precedent protecting creative workers against generative AI.

  11. Generative mediaReach 39

    Getty's legally safe AI image generator

    Getty Images just teamed up with Nvidia to launch Generative AI by Getty Images, an image generator trained strictly on Getty’s proprietary licensed library. The tool offers full copyright indemnification for commercial users and blocks prompts featuring real-world figures or specific artist styles. Why it matters: As copyright lawsuits mount across the industry, Getty is betting that legally bulletproof images will win over cautious enterprise clients.

  12. AI toolsReach 31

    UPS deploys AI to fight porch pirates

    UPS just expanded DeliveryDefense, an AI tool designed to predict which doorstep deliveries are likely to get stolen. The model crunches billions of delivery records to score addresses from 1 to 1,000, allowing merchants to reroute sketchy drops to pickup lockers before theft occurs. With holiday theft peaking, UPS is betting algorithmic risk scoring beats doorbell cameras.

  13. IndustryReach 56

    George R.R. Martin and bestselling authors sue OpenAI

    The Authors Guild and 17 top authors, including George R.R. Martin and John Grisham, just filed a copyright infringement lawsuit against OpenAI. The complaint alleges OpenAI copied their books wholesale to train LLMs without permission or licensing fees. OpenAI responded that it is actively in productive talks with creators.

  14. IndustryReach 50

    Roblox buys AI voice startup Speechly to clean up voice chat

    Roblox just acquired Finnish voice tech startup Speechly to tackle toxic behavior in voice chat across its platform. Speechly builds AI tools for real-time speech moderation and transcription. The move will help Roblox protect users as it rolls out avatar-based voice calls, though financial terms remain undisclosed.

  15. IndustryReach 40

    TikTok wants creators to label their AI videos

    TikTok just launched a new toolset that lets creators label posts made or enhanced by AI. Disclosures are now mandatory for realistic synthetic media, existing in-app filters are being renamed to include "AI," and the company is already cooking up an algorithm to detect and label AI posts automatically.

  16. IndustryReach 37

    OpenAI builds a dedicated red team network to stress-test models

    OpenAI just launched the Red Teaming Network, calling on domain experts worldwide to rigorously stress-test and evaluate its AI models. Members are compensated for their work under NDAs and can commit as little as 5 to 10 hours a year. Why it matters: It shifts OpenAI's model evaluations from one-off audits into a continuous, diverse collaborative defense.

  17. IndustryReach 43

    Microsoft AI researchers accidentally expose 38TB of internal data

    Microsoft's AI research team accidentally leaked 38 terabytes of private data—including secret keys, credentials, and 30,000 internal Teams messages—via an overly permissive Azure link on GitHub. Cloud security firm Wiz uncovered the slipup, which mistakenly granted full control over the storage account since 2020. Microsoft has revoked the token and upgraded its secret-scanning safeguards.

  18. IndustryReach 37

    Hollywood stars turn to Metaphysic to guard their digital clones

    Metaphysic just launched Metaphysic Pro, a new platform designed to help actors create and manage their own AI-generated likeness and voice data. Early adopters include Tom Hanks and Anne Hathaway. Why it matters: As Hollywood strikes over AI rights, performers are actively looking for ways to lock down their consent and get paid when their digital twins work.

  19. IndustryReach 35

    Mark Zuckerberg defends open-source AI at Senate forum

    Mark Zuckerberg just laid out Meta's AI vision at a Senate forum, framing safety and access as the technology's two defining challenges. He defended open models like Llama 2 as crucial for American competitiveness and security, though Meta will rethink that stance if systems ever approach superintelligence.

  20. IndustryReach 40

    Pulitzer-winning authors are suing OpenAI again

    OpenAI just got hit with another class-action lawsuit from acclaimed authors including Michael Chabon and David Henry Hwang. The suit claims OpenAI trained GPT models on their copyrighted books without permission or payment, arguing the chatbot couldn't summarize their works otherwise. Why it matters: Legal scrutiny around training data is rapidly snowballing, and creators want their cut.