Jesse Wynants’ Post

A "customer" thanked support for such a fast reply. It was a bot. 🤖 As of today, August 2, that goes against the rules if the company doesn't say so. Article 50 of the EU AI Act kicks in today, and it's not really new information, it's just finally enforceable. The core ask: 🤖 AI chatbots and voice agents must tell people they're AI, right at the start of the interaction 👉 AI-generated images, audio, and video need a machine-readable mark showing they're synthetic 🤖 Deepfakes and AI-written news content need a clear label, no exceptions for "it was obvious" Fines go up to €15M or 3% of global turnover, whichever is bigger. And it doesn't matter where the company is based. If EU users are on the other end, the rule applies. Here's the part marketers keep missing: this isn't really about penalizing AI use. It's about killing the illusion of a human that isn't there. A support bot that says "I'm an AI assistant" isn't weaker than one pretending to be Sarah from the team. People have gotten pretty comfortable with AI. What burns trust is finding out later they were lied to about it. Curious how others are reading this: compliance headache, or the thing that finally forces good practice?

We put a “robust” AI competitor to the test. It didn’t just fail, it collapsed. In minutes, a user got it to: - Confidently give wrong or outdated info - Draft fake legal filings against its own org - Agree its leadership was incompetent - Share a playbook for manipulating the company Your event needs a business problem solver, not a risky pattern generator. https://hubs.la/Q04lzRTD0

AI isn’t a tool to shy away from anymore, especially in public relations. An AI Agent is an autonomous system that takes a goal and plans the steps to achieve it, execute the campaign and observe outcomes. See how this tool is shaping PR: https://lnkd.in/eKnn33HK #PR #LaTorre #AI

More people are skipping the search bar and just asking their AI assistant directly — "who's good for X," "what should I use for Y" the same way they'd ask a friend. Except the friend is ChatGPT, Copilot, or Gemini. That's a different kind of visibility than SEO. Ranking #1 on Google doesn't matter if the assistant your customer actually opened has never heard of your business and most businesses have no way to know if that's the case. One thing you can check yourself in 10 seconds: go to yoursite.com/robots.txt. If you see GPTBot, ClaudeBot, or PerplexityBot listed under "Disallow," you've blocked the exact bots that let an AI learn about your business — often without knowing it (a leftover default from an old plugin or host setting). That's one of eight things I check in the AI Visibility Audit I just added. Drop your URL in the comments and I'll tell you what I find 👇 andyfreelancer.com

Everyone's talking about Claude adding watermarks to AI content this week. So I built a tool that actually reads them. Most "AI detectors" give you a probability score. This one reads the provenance data embedded in the file at generation time. The cryptographic signature, the signer identity, the soft-binding watermark that survives metadata stripping and screenshots. Here's what it detects right now: 🖼️ Images (Claude, Gemini, DALL·E, Firefly, Midjourney, Grok) C2PA cryptographic signature and signer identity ("Google C2PA Media Services" shows up on real Gemini images) TrustMark soft-binding watermark, which reads the invisible pixel-level signal even after all metadata is stripped or the image is screenshotted Perceptual fingerprint for similarity tracking 📝 Text Invisible Unicode characters including zero-width spaces, bidi controls, and tag-channel steganography Formatting fingerprints - narrow no-break spaces are a consistent ChatGPT tell Stylometric estimate that shows its reasoning 🎵🎬 Audio and video Same C2PA pipeline, modality doesn't matter On Claude's statistical text watermark: it works at the token-generation level and requires Anthropic's private key to verify. I have the integration built and ready. It goes live the day Anthropic opens the API. No login. No fake confidence scores. A clean result doesn't mean human-made. 👉 https://lnkd.in/gJTur4Kn

Starting August 2, the EU requires chatbots to tell users they're talking to AI. AI-generated audio, images, video, and text must carry machine-readable marks so automated systems can detect them. Deepfakes need visible labels. Providers who ignore these obligations face fines up to €15 million or 3% of worldwide annual turnover, whichever is higher. The rules apply under Article 50 of the EU AI Act to any company serving EU customers or employees, covering 450 million people. That includes every major AI provider operating at scale, from OpenAI and Anthropic to Google and hundreds of smaller platforms. Over 180 organizations signed the EU's voluntary Code of Practice on transparency, giving companies a recognized compliance path. Signing doesn't exempt them from the underlying legal obligation, but it documents their approach for regulators. The AI Act's other rules move on different timelines. High-risk AI system requirements land December 2, 2027. New prohibitions on non-consensual explicit content arrive December 2, 2026. The chatbot disclosure and content marking requirements are already in effect. If your company runs a chatbot, voice assistant, or any AI system interacting directly with people in the EU, those disclosure obligations are active today. https://lnkd.in/eXxwGqHg

A shareable AI link is not the same thing as a private handoff. Recent reports found publicly shared Claude conversations, apps, documents and other AI-created artifacts appearing in Google and Bing search results. These conversations were private by default. The risk changed when someone generated a public link and then shared that link somewhere a search engine could find it. In some cases, people likely thought they were simply sending something to a colleague, friend or client. They may not have realized they were creating a public web page. ChatGPT experienced a similar controversy in 2025 and removed the option that allowed users to make shared conversations discoverable by search engines. But its shared links are still accessible to anyone who receives the URL, and that person can share it with someone else. That is the reminder here: when you create a shareable link, you are creating a live digital artifact. It could be a conversation, a website, an application, an animation, a document, a prototype or an AI-generated tool. It may look temporary, but it now exists on someone else’s infrastructure and may be copied, forwarded, archived or discovered in ways you did not originally anticipate. The platforms also have a responsibility to use privacy-protective defaults and technical controls such as “noindex.” But we should still practise some basic digital housekeeping. Take a few minutes to review the shared content in the AI tools you use. Remove links that are no longer needed. Check whether old prototypes and public sites are still online. Review what information was included before you shared them. Never assume that “anyone with the link” means “only the person I sent it to.” For Claude, shared chats can be reviewed under Settings → Privacy → Shared chats. For ChatGPT, they can be reviewed under Settings → Data Controls → Shared links. You do not need to check these settings every week. But you should look at them at least once, understand what they control and return periodically to clean up. It is similar to buying a car. You can start driving without reading every page of the manual, but it is still worth learning what the controls do before you need them. AI literacy is not only about writing better prompts or generating more impressive work. It is also about understanding what happens to that work after you click Share. AI drafts. Humans decide. Humans should also know where those drafts are going. CivicPlay.ai 🍁 #AIPrivacy #ResponsibleAI #AIGovernance #DigitalLiteracy #CyberSecurity

Most MSPs I talk to are adding AI to their pitch deck. Few have thought through what they're actually delivering. A managed AI service isn't just "access to ChatGPT with a markup." If it were, your clients would do it themselves. Here's what separates a real managed AI offer from a rushed one: ✅ You control which models clients can use - not just "AI," but GPT, Claude. You decide what's in scope. ✅ You set the guardrails before day one - data policies, prompt rules, what can and can't leave the environment. ✅ You can answer the question every client will ask - 'What is our AI actually doing?' You have the data. You have the report. ✅ You own the onboarding - not a vendor webinar. You. With context about their team, their risks, their workflows. ✅ You have a renewal conversation - not just a licence key. A review, a report, a reason to stay. If you can check all five, you're selling a service. If you can't, you're reselling a tool. Which one are you building?

The most significant shift in AI adoption is not happening in the boardroom. It is happening at the freemium tier. OpenAI announced this week that unlimited text chats are coming to ChatGPT free users, backed by a new GPT-5.6 Luna model. The company now counts over one billion weekly active users. When a platform crosses that threshold and then removes usage ceilings for its free tier, it signals something bigger than a product update: it signals that the economics of inference have changed enough to make scarcity irrelevant. This matters enormously for how businesses think about AI strategy. For years, the conversation centered on which company had the best model. That race is far from over, but a parallel race is now underway: who can deliver AI at utility scale, to everyone, at near-zero marginal cost. The freemium tier is no longer a loss leader for conversions; it is a distribution infrastructure that shapes default behaviors, expectations, and workflows across entire populations. Organizations that are still piloting AI tools internally, waiting for the right moment to scale, should note what this moment is: the baseline expectation for what AI can do, for free, just moved up considerably. https://lnkd.in/dg_9ws7W

What is Article 50 of the EU AI Act? Article 50 is the transparency rule. It's the part of the Act that says people need to know when they're looking at, listening to, or talking to something AI made or generated. It's been enforceable since 2 August 2026, and it wasn't pushed back by the Digital Omnibus the way the heavier high-risk rules were. It covers a few specific situations. Content that's a deepfake, a realistic image, video, or audio of a real person doing or saying something they didn't. AI chatbots, where people need to know they're talking to a machine, not a human. And AI-generated text published to inform the public on matters of public interest, which needs disclosure unless a human has reviewed and taken editorial responsibility for it. Take a beauty brand running a customer service chatbot on Instagram DMs. Someone messages asking about a return, gets a reply within seconds, assumes a staff member's online late. That's exactly the situation Article 50 is built for. The brand needs to make it clear, somewhere in that interaction, that they're talking to an AI assistant. Doesn't need to be dramatic. A line in the opening message does the job. Breach it, and fines go up to fifteen million euros or three percent of global turnover, whichever's higher for larger companies, whichever's lower for SMEs.