Marko Dinic

On a big public records request, I think the slow step is often building the search, before anything even runs. A request can involve nicknames, initials, misspelled names, and a scanned lease, with records spread across email, texts, WhatsApp, and social media. The query you build should also be the record of how you searched. Jatheon Cloud puts every name in one keyword list, reads scanned files with OCR, and logs every search, so you can show the requester exactly what ran. jatheon.com/blog/keyword-sea…

WWE’s lawyers told Vince McMahon and four other execs to keep their messages. Four of them set Signal to auto-delete anyway, one on the same day the Justice Department asked for information. When shareholders sued over the UFC deal, the chats were gone, so the judge assumed the worst. The defendants agreed to pay $147.5 million a few days before trial. I think a hold notice alone doesn’t do much when people can delete on their own phones. Work chats should be archived automatically, with the hold on the archived copy, so the messages are still there when a judge asks.

Phone storage ends up deciding how long Sonoma County officials keep their texts. The county’s records retention schedule leaves texts out. Three of the five supervisors say they delete theirs regularly to free up memory. California’s Supreme Court ruled in 2017 that texts about public business can be public records, even on a personal phone. The First Amendment Coalition, a transparency nonprofit, recommends keeping all local government correspondence for two years. I’d capture texts automatically and keep them that long. That’s what Jatheon does, on iPhones and Android phones. jatheon.com/products/text-me…

The Heppner ruling was about privilege, and the lesson for companies goes further. AI chats are records now. Heppner, a former CEO, used the consumer version of Claude to plan his fraud defense. The judge ruled the chats weren’t privileged. Claude isn’t a lawyer, and Anthropic’s consumer terms said it could share what users type with the government. Prosecutors got every one, and a jury convicted him. Your employees ask Claude about contracts and HR issues every day, often from personal accounts you don’t control. When a regulator or opposing counsel asks for those chats, you’re expected to produce them. The IT and compliance leaders I talk to know their teams use AI at work. Very few could pull a single chat tomorrow. We archive Claude now. Chats are retained by policy, searchable next to email and Slack, and ready for legal hold. jatheon.com/solutions/commun…

Next Tuesday, on a live webinar, somebody on my team will forward a @WhatsApp message that should never have left the building, on purpose. We build the content policy first, on screen, then trigger it. The message disappears from the employee’s phone and the original stays in the archive, with its timestamp. That’s one of the things Jeff Marlow and I will be showing live in @Jatheon Cloud on September 29 at 2 pm ET. You’ll also see the @SlackHQ archiving connector, how @claudeai capture works and how we score emails for risk with AI. This will be a product-focused session end to end, and we can also show you how to use the platform for your specific use case. If you can’t make it live, register anyway and you’ll get the recording. Register: luma.com/zzldyuap

The SEC has been reading advisers’ websites. Twenty-three US firms have paid $2.29 million since 2023 for what it found there. Hypothetical returns shown to anyone who landed on the page, and two firms with no copy of their own ads. Promotional pages are advertisements under the Marketing Rule, and a separate records rule says keep a copy of every ad you put out. A website also changes every week. Screenshots don’t prove what a page said on a given day, and a US appeals court has already refused Wayback Machine snapshots as evidence without someone to vouch for them. I’d archive the website like email, every version with its date. hubs.la/Q04xK2d80

In the US, anyone can ask a federal agency for its records, and last year that happened 1.7 million times. A complicated request at the State Department now takes 270 working days, against the 20 the law allows. Canada runs the same idea on a 30 day clock, and an institution can extend itself for as long as it calls reasonable. At Global Affairs, 35% of requests took over a year. Miss the deadline in the US and the next step is a lawsuit. In Canada it is a free complaint to the Information Commissioner, 375 binding orders last year, some ignored until Federal Court. My team broke down the nine reasons a US agency can legally refuse a record. jatheon.com/blog/foia-reques…

Anthropic kept a 30-day copy of every business prompt since June, and this week it gave in to regulated customers and stopped. OpenAI made the same promise two weeks ago, so the only copy of a work chat with ChatGPT that can still exist is on your side. That is good news for privacy. It also means that when an auditor or a lawyer asks for the conversation behind a decision, the vendor truthfully says it has nothing. If nobody on your side saved it, that conversation never happened as far as the record goes. The job of remembering just moved from their servers to yours, and I like it. anthropic.com/news/enterpris…

Reporters in Washington state pulled thousands of pages of city officials’ ChatGPT logs through public records requests. The same laws cover a school district’s ChatGPT history, and I’d bet staff are already pasting student details into these tools. That’s student data outside district systems, and records you can’t produce when a request lands. You can’t retain what you never captured. My team’s K-12 mandate map now has a line for AI chat, and that's the one I'd check first.  bit.ly/4x72eRs

FINRA fined Benjamin F. Edwards $750,000 in January over business texts it never captured, a full year after the SEC ran its last texting sweep and moved on. Chairman Atkins now calls that sweep “exactly how regulators should not act,” and he has a point, it never produced a case of investor harm. But the sweep ending isn’t the rule ending. SEC 17a-4, Advisers Act 204-2 and FINRA 4511 are all still there, and FINRA runs its own book. Texting clients is allowed. The rules ask that a business text be captured, retained and reviewable, same as email. We sell archiving software, so weigh this accordingly, but the rule-by-rule guide is free and ungated. jatheon.com/blog/text-messag…

Did you know document review costs about $14,000 per gigabyte for an average legal case? RAND, the think tank, did the math back in 2012, and lawyer rates only went up since. Lawyers bill by the hour, so the bill comes down to how much data you hand them. Five copies of every email (Exchange does not deduplicate), customer chats in WhatsApp that never got captured, a thousand voice mails without transcriptions. Doesn’t matter what the excuse is, you pay for all of it at the highest hourly rate. By the time a lawsuit lands, you are scrambling to meet discovery deadlines and the data is what it is. My team put the whole document review process in one guide, including the archiving setup that will keep your bill down. jatheon.com/blog/compliance-…

Microsoft Teams starts keeping AI archives of meetings in August, by default. A saved transcript plus one Copilot license in the room gets you a .meeting file in tenant SharePoint, kept five years by default. The idea makes sense, organizations forget what they decide. But it also gives Microsoft detailed information about your business, and five years is a long time to share your business decisions. Partners this close can turn into competitors, look at Anthropic: its product chief sat on Figma’s board, and three days after he stepped down Anthropic released Claude Design, a direct competitor. Microsoft calls the policy AI Memory and Archive, and turning it off is a manual admin step. New record types usually arrive because a regulator asked. This one arrives as a default, with a retention period your org didn’t choose. We are seeing more demand for archiving AI conversations across the board, so it may be timely. Just buyer beware. The setting path and the rest of the August changes: windowslatest.com/2026/07/17…

Monzo approved customers who gave “Buckingham Palace” and “10 Downing Street” as their home addresses. Its own policy said UK residents only. Technically, the palace qualifies. The same bank promised its regulator in writing to stop taking on high-risk customers, then opened 33,039 accounts for exactly that group. The FCA’s bill came to £21 million, three years after the failures ended. The FCA fines retroactively, and it judges a bank against its own written policy. You answer years from now for the rules you write this quarter. My team mapped the six areas the FCA examines and what each expects in one guide. jatheon.com/blog/fca-complia…

The biggest AI procurement question over the next few years won’t be “Which model performs best?” I think it will be “Who controls whether you can keep using it?” China launched a 29-country AI cooperation organization in Shanghai this week. Three weeks earlier, the US brought together 35 countries for its own AI partnership in Washington. China’s message wasn’t about innovation only, they’re offering developing countries open AI models and shared AI infrastructure. I don’t see this as charity, open models are China’s distribution strategy. The US has taken the opposite approach, its most advanced chips and frontier models sit behind export controls and licensing restrictions. China is offering access. Enterprise software is no longer just software, it comes with an AI model, a country of origin, and a rulebook attached. Before my team buys an AI-enabled platform, I’d want to know where the model was developed and which government can restrict access to it. We’ve already seen how quickly this plays out with Fable 5. If access disappears after you’ve built critical workflows around a model, you’re cooked. Overnight. reuters.com/world/asia-pacif…

European Union has zero leading AI companies but thinks it knows how to govern AI. Its AI Act deadline for “high-risk systems” just moved to December 2027. The reason is unflattering, the standards that spell out what compliant looks like still aren’t published. Once they’re established they’re the benchmark, and that deadline can arrive early but never late. We sell the system that helps organizations comply, and we need to comply ourselves too. Either 1) wait for the final government specs and get our ducks in order under pressure and a deadline, or 2) start the classification, documentation, and evidence trail now on a blind guess, then adjust when the specs land. A “high-risk system” has to show what data it used and what it decided. When the auditor asks, there’s no time to collect that from Exchange, Teams, Slack, and backups, only time to search it. Jatheon compiles all your communication data, preserves chain of custody, and makes it easy to search, old school or with AI. I still think this is the cheapest time you’ll get to wire it up - stress free. digital-strategy.ec.europa.e…

Almost all B2B SaaS software pricing is a black box. Sometimes on purpose, other times since the pricing is complex due to the modular nature of enterprise systems. We wanted to make email archiving pricing and gotcha a bit more transparent, so here it is. 22 years of experience in archiving distilled into a buyer's guide including the full cost breakdown, oddities and questions I’d ask any vendor, us included. jatheon.com/blog/email-archi…

84% of European IT teams are confident their collaboration tools are secure. Only 29% of them say those tools are fit for sensitive data. A third can’t even say who inside their own company can see sensitive files. And when getting an outside partner access takes a ticket and a wait, the sensitive stuff moves to WhatsApp or a personal Dropbox in seconds. What I keep coming back to is that none of this shows up until someone asks for proof. The confidence rarely survives the first audit that wants specifics. bit.ly/4wxStLq

OpenAI spent last month sitting on GPT-5.6, its own finished product, waiting for the government to clear it for release. That’s what frontier AI is now: a controlled technology, closer to weapons than to ordinary software. No vote, no rulebook. One June executive order lets the government spend 30 days inside any model it calls powerful enough before you can ship it, and it’s still called voluntary. Anthropic got it harder. Its two strongest models were switched off overnight under export controls, the same lever used for weapons. The one built to find software vulnerabilities is still locked down. Two of the strongest AI labs in the country, both made to ask permission, judged against a rule that doesn’t exist yet. I get the caution. I just wish they’d finish the rulebook before enforcing it.