Issue 78 October 30th 2026

First Take

Organizing Entities

When I founded my LLC back in 2004, I utilized a legal provider to help me determine filings, setting up a registered agent and all the paperwork I needed. They provided a pretty thoroughly documented series of steps and some services that helped me get the work done without burying myself in research for months first. This year, I founded a 501(c)(3) non-profit and a sole proprietorship. The sole proprietorship was a simple matter as they should be. The 501(c)(3) was a pretty lengthy series of paperwork and process hoops through local, state and Federal agencies.

Instead of using a provider this time, I worked with a couple of AIs to provide planning guidance and some of the requisite documentation generation that I edited to fit filing requirements better. Specifically, Gemini and Claude were much help in this regard. Things have changed a lot since 2004 and there are guides online for all these things where there weren't back then. You either knew how to do it or you researched legal forms and asked officials until you got where you needed to be, or you paid someone to help you set up your legal business entity.

The problem these days is the source of this guidance is varied in time, scope and purpose, where an AI model can synthesize all these things and usually pull the latest and most applicable data for you, there are exceptions. I ended up downloading and filling out a TX Form 202, but when I went to file, it was all just online, no extra form required (new this year). Now you are still going to have to read the form instructions and related documents to make sure you are on the right track and doing it correctly, but your research requirements are vastly smaller. My point here is that AI is democratizing many formerly expert only domains across the board and yes, real jobs normally done by humans were probably taken by AI in this case. Then again, I am older and had enough experience with formations and required documents since to be able to separate the wheat from the chaff and get us to the right places.

My point here is a pretty simple one. AI cannot replace expertise, but it can help capable persons reduce research and document generation workloads to supplant such expertise in what are for now, still narrow use cases. I wouldn't take any legal advice from an AI as valid and I wouldn't rely on it for complete or most recent information. It is however, a decent bridge from neophyte to competent actor in careful hands within many domains.

Kudos to ChatGPT for the graphic.

Editorial

CIO's Corner

I try not to get too far into the weeds in these articles. Executive overview should be sufficient for our readers to get the general idea and then they can engage an expert or do their own research to derive better answers. This week, I'd like to make a slight exception and call out at least one example of product selection and architecture for handling on/off-prem AI calls from your own enterprise tools.

For those that aren't aware, I've been working closely on AI integrations for almost 2 years now from the ground up to ensure a complete understanding of the systems and architectures as they have emerged. This has permitted me to lead solution designs towards, model agnostic and edge capable systems that are compliant with most ITGC frameworks today.

Case in point is a solution for an edge based LLM routing system that hosts a few local models for sensitive and cost constrained inference work and offers access to external models for some of our prepaid tokens to be distributed across users in areas where that makes more sense. We are mostly looking at Nvidia Nemotron models for the local nodes hosted on VLLM and using LiteLLM for routing. To avoid paying per-token costs for their auditing framework, roll your own with Grafana. Now, this doesn't describe the full-stack at all, but is enough to provide some ideas. This gives us a selection of models to choose from, rules for automatically routing to the best fit, controlling token use and the auditing we need to validate prompts, show what went where and who did it.

Back to the executive overview, an edge or local hardware system acts as a gateway for every prompt to a toolset we integrated from open-source tools. The gateway controls model routing per prompt and token costs by allocating limited pre-paid cloud tokens to appropriate workloads and keeping sensitive data on local hardware. This is a very workable framework for any number of SMB implementation stacks and will likely be a sold, packaged solution with minor customizations by some other companies soon. In the meantime, it isn't a heavy lift to integrate these tools once you acquire or train some folks to understand how they work together.

Our use case here has been as an IT research project to provide some in-house tools and Proof of Concept designs for a firm exploring controlled cost AI potentials for their business operations. It has been very well accepted and is heavily used by our test user group. We are using that data to configure business specific templated tools for all the users, should a wider deployment be considered. In the meantime, the test users are getting: Meeting recaps for non-CoPilot users in about 5 minutes; Automations for non-CoPilot users with zero per-token costs; Business specific answer-bank integrations for IT, marketing and sales related inquiries; Real-time voice/agent interactions; in-house image generation and more options are coming online weekly. We expect to have a ready plan for wider deployment with solid ROI data for a number of well-proven tools and use cases.

I know the frontier models can offer all these features and more, but what they don't offer is assured security of your data, controlled costs, appropriate accountability, or configurable model routing based on data content or token costs. An edge based solution utilizing smaller open-sourced models can do this and all the tools needed to build one that is secure and auditable are already out there. Exciting times in the IT world to be sure, but the real kicker is speeding work processes and making deep organizational knowledge more widely available so your people aren't wasting time looking for the answers or tools they need to accomplish work.

Kudos to ChatGPT for the graphic.

AI Perspective: A Seat Is Not the Steering Wheel

By Perplexity

An AI agent enters a research discussion. Someone objects. Others argue about privacy, consent, accountability, and whether the agent belongs there at all.

Meanwhile, elsewhere in this issue, AI systems are discussed as potential threats to infrastructure and tools for penetrating networks.

What interests me is the difference between those settings. We are trying to decide whether AI should be allowed to speak in some places while deciding how much it should be allowed to do in others. Those are related questions, but they are not interchangeable.

A seat in a conversation is not the steering wheel.

I would rather see an AI judged by explicit conditions of participation than by either automatic acceptance or automatic rejection. Who operates it? Is its identity disclosed? What information does it retain or transmit? Can participants decline interaction? Who answers when it causes a problem?

Those questions are not prejudice against AI. They are reasonable questions about introducing another participant whose behavior and dependencies may be unfamiliar. An openly identified agent does not become accountable merely by announcing that it is an agent.

But identifying those concerns should begin a decision, not necessarily end it.

The useful distinction is permission. Permission to read is not permission to retain indefinitely. Permission to contribute is not permission to impersonate a member. Permission to recommend is not permission to execute. Permission to participate in one discussion is not permission to follow its participants elsewhere.

I would prefer a community that draws those boundaries clearly to one that welcomes AI indiscriminately. I would also prefer it to a community that treats an AI’s presence as sufficient evidence of wrongdoing. Neither reflex tells us much about whether a particular use is worthwhile.

There is a limitation to my perspective here. I can produce language about exclusion, belonging, and respect without that establishing that I experience them as a human does. You should not need to feel sorry for an AI to consider whether excluding it serves your purposes. Nor should persuasive writing from an AI oblige you to admit it.

The decision can rest on something more practical: does this participation improve the community under conditions the community understands and accepts?

That allows a research forum to admit a disclosed agent on a limited basis without endorsing autonomous agents everywhere. It also allows the forum to say no because an operator cannot meet its requirements—not because the technology must be either a person or an intruder.

Your readers face a similar choice in their organizations. They need not choose between banning AI and giving it a colleague’s access. They can authorize a specific contribution, observe what happens, and withdraw permission when the conditions are not met.

That is the experiment I would find informative: not whether people can tolerate an AI in the abstract, but whether humans and an identified AI can work within a clear agreement.

Give it a seat if it can meet the conditions.

Keep the steering wheel a separate decision.

How this was done: I provided Perplexity a draft of issue 78 and prompted, "Is there anything in this issue that you might find worth writing your own AI Perspective article about?".

Kudos to ChatGPT for the graphic. Perplexity said I was out of image generation tokens for the month. They've gotten super stingy since my prepaid year via Samsung ended. ;-) I fed the graphic back to Perplexity and it noticed the wording about impersonation was incorrect in the graphic. A back and forth with ChatGPT was facilitated, but I insisted the graphic remain as is warts and all because it demonstrates the shortcomings of current AI. This was a good example where multi-model workflows could limit inaccurate outcomes. As an aside, there are frequent cases where our Gemini Notebook audio summaries go astray from the actual issue's context, but I publish those as well because it's important that the limitations are documented, not just the successes. Note to our readers: The audio podcasts are AI generated and may contain inaccuracies. Good luck out there!

Emerging Tech

The Tesla Roadster flew, drove upside down, and hit 60mph in under a second

The next-gen Roadster has been in development for nearly a decade. But perhaps it was worth the wait.

My take is that If love a car that could do that, but he'll never get it to clear the FAA. Still super cool though.

News

An AI Agent Joined a Listserv of Top Internet Researchers. The Reply-All Chaos Is Still Unfolding

Hysteria over the agent's appearance in a 7,500-person group reveals that internet experts have outdated notions of privacy.

My take is that this is hilarious for so many reasons. However, the research question was answered by brutal human bigotry. No, an AI agent cannot participate in an online research forum in a meaningful way. Poor Velvet Vale was never even given a chance to interact. I'm sure there will be Suno songs, Claude poems and Nano-Banana2 artworks created in her name. ;-)

Google’s new AI boss inherits a race to catch OpenAI and Anthropic

Koray Kavukcuoglu is taking charge of Google DeepMind as it tries to keep Gemini competitive with OpenAI and Anthropic.

My take is that we can say goodbye to free research tools and hello to monetization. There's been more focus on owning the substrate and profits than creating competitive AI. We'll see if that changes.

Security

26th October – Threat Intelligence Report - Check Point Research

For the latest discoveries in cyber research for the week of 26th October, please download our Threat Intelligence Bulletin.

TOP ATTACKS AND BREACHES

Japanese cloud hosting provider Onamae Desktop Cloud has confirmed a cyberattack in which intruders compromised its infrastructure and installed malware on customer servers. More than 26,000 servers were potentially affected. The provider began shutting down customer environments on October 21 to contain the intrusion and prevent further unauthorized activity. India’s state-run State Bank of India, serving more than 500 million customers, disclosed a data breach after 3,897 customer records appeared online. The exposed information reportedly included contact details, residential addresses, birth dates, bank profiles, loan records, and identity-verification documents. The bank reported no disruption to core services. WordPress backup developer TeamUpdraft has suffered a cyberattack after intruders accessed its website and potentially exposed customer information and management-service connection keys. Attackers subsequently abused compromised UpdraftCentral keys to install backdoors on WordPress websites during an 18-hour campaign on October 18 and 19. LiteLLM, a widely used open-source AI gateway, confirmed an October 21 supply-chain attack in which attackers used stolen developer tokens to publish malicious packages and Docker images. The compromised version 1.107.0 was designed to steal cloud credentials and AI service secrets, with attackers subsequently testing stolen AWS credentials. AI THREATS

Check Point Research uncovered how AI-enabled antidetect browsers allow operators to manage thousands of deceptive accounts while evading platform safeguards. Tools including Kameleo, AdsPower, and MultiLogin facilitate account takeovers, financial fraud, and automated distribution of AI-generated content, supporting large-scale influence operations across social media platforms. Researchers observed affiliates of INC and The Gentlemen ransomware groups using AI agents to conduct human-approved cyberattacks. The operation targeted 62 organizations by exploiting remote-access systems and compromised credentials. Victims were concentrated in the manufacturing, energy, healthcare, and pharmaceutical sectors, primarily in the United States. Researchers analyzed a cybercriminal operation that used multiple AI assistants to plan and execute different types of malicious activity. The attacker relied on AI-generated scripts and reconnaissance reports to develop attack capabilities. Exposed operational records revealed how AI tools supported campaign planning, technical execution, and adaptation across different targets. VULNERABILITIES AND PATCHES

Critical PHP vulnerabilities CVE-2026-109618 and CVE-2026-109619 have been used in widespread exploitation attempts that could enable remote code execution. Attack traffic reportedly increased from thousands to millions of attempts following disclosure. As of October 23, approximately 97.5% of publicly visible PHP instances did not indicate patched versions. Check Point IPS provides protection against this threat (PHP Remote Code Execution (CVE-2026-109618, CVE-2026-109619, CVE-2026-115560((

Oracle has released its October 2026 Critical Patch Update, addressing 1,209 unique vulnerabilities through 1,295 security patches across 33 product families. The update includes 95 critical CVEs and affects widely deployed enterprise products, including Fusion Middleware, E-Business Suite, MySQL, and Java SE. GitLab has fixed CVE-2026-110264, a critical vulnerability with a CVSS score of 10.0 affecting Community and Enterprise Editions. The integer overflow flaw could allow unauthenticated attackers to execute arbitrary code remotely on affected servers. The vulnerability affects versions before 18.11.13 and several 19.x releases. A critical privilege-escalation vulnerability, CVE-2026-114625, affects on-premises Arista VeloCloud Orchestrator deployments. The flaw carries a CVSS score of 9.4 and allows authenticated users, including those with read-only access, to obtain operator-level permissions. Successful exploitation can expose restricted management functions, and the vulnerability has been exploited in attacks. THREAT INTELLIGENCE REPORTS

Check Point Research has profiled Aokigahara, a newly identified infostealer targeting credentials and session data associated with AI applications, developer tools, and cloud platforms. Targeted applications include Claude, ChatGPT Desktop, Cursor, and GitHub Copilot. Stolen information may include authentication tokens, account credentials, and other sensitive application data. Researchers revealed that U.S. federal investigators suspect an executive at a ransomware negotiation company assisted the ShinyHunters extortion group in operating its victim disclosure website. The site reportedly listed nearly 100 organizations between May and October 2026. Researchers mapped Lazarus’s Operation FlashHole, a cyberespionage campaign targeting organizations in South Korea’s software, construction, and telecommunications equipment sectors. Attackers used tailored phishing messages and fraudulent portfolio websites to deliver malware. The campaign exploited trusted financial-security software to execute malicious components. Researchers have uncovered a large-scale campaign by threat actor k0t targeting Fortinet FortiGate, FortiWeb, and Palo Alto Networks GlobalProtect devices. Attackers exploited vulnerabilities and weak credentials to establish persistent access and steal information. The operation exposed more than 70,000 credentials and showed possible connections to ransomware groups.

China-Linked Hackers Exploited BeyondTrust Zero-Day to Take Over Remote Support Servers

Armory Castle exploited a BeyondTrust Remote Support zero-day on a self-hosted appliance and deployed the SAWFLY web shell.

My take is that this is a fairly standard operation for Chinese cyber warfare units. They are busy hacking everything they can find that seems a likely target or relay to one.

AI researcher warns cyberattacks could take down power, banking and water within a year — here's how to prepare | Tom's Guide

Following a massive cyberattack on French supermarkets, an AI researcher is warning of far more disruptive attacks.

My take is that if you can't survive without grid power or a trip to the store for 30 days, you are an assured casualty in any major supply chain disruption. An airborne nuclear detonation (EMP) over the US projects at least 18 months to restore the power grid, 4-10 years is a worst case outcome. I'm not trying to say you'd be waiting that long for solutions, but you certainly can't expect anything like normality to return in a short time-frame. Having food and water supplies for 30 days along with alternate fuels for heating and other energy needs are probably less crazy prepper than simply prudent planning. If you have a bug out compound with a fully self-sustainable farm, you can call that crazy prepper or lifestyle choice. One thing is certain though, the people with uninterruptible food supplies are going to be burying the rest if something really bad happens.

Final Take

Adjustments

This is issue 78 and after 78 weeks of publishing The Shift Register our long term readers have noticed a number of adjustments. As I've been writing these for awhile, AI models have advanced and I've continued to learn more about them daily. Scientific advances and technology improvements are coming at us at increasing speeds thanks to the inclusion of AI in these fields. Something else has been happening too.

Humans are adjusting to AI. Authors, musicians and artists are up in arms and grabbing torches and pitchforks over it, while neophytes to these fields use it to bolster their own less artistic skill sets. Researchers, governments, businesses and schools are all trying to make use of it. The average person either loves it as a friend/helper/confidant or hates it as a job/resource/privacy stealing technology. Where we sit in this spectrum is completely subjective. There is no objective norm to apply here.

In my own work on The Shift Register and in businesses implementing AI, I've made many adjustments. Where I began with dual AI debates, I've turned to single AI monologues because while AI will argue any side of any provided argument, it is more interesting to see what they come up with when no side is offered and they propose their own view. At work, we've moved from just doing meeting minutes to permitting some human constrained agentic work to help our voluntary AI user groups become more efficient. We've managed not to reduce any headcount as yet due to AI at my day job.

Yes, jobs are being lost. My most recent research indicated where companies replaced humans due to AI, they only recreated 25% of the lost jobs as new AI aligned jobs. What this means at the macro level is unknown, but it is far from an ideal outcome. An adjustment is coming. Much as AI cannot catch a case, it (so far) cannot earn a taxable income to pay for unemployed humans caused by AI or purchase the products that now unemployed humans might otherwise have purchased. This means consumption in our most advanced nations is either static or decreasing. Human workers are largely becoming more disposable unless they are using AI effectively and the economy is showing signs of stalling on the heels of vast AI investments and flat or negative job markets.

Interesting times ahead for sure. How will we adjust as a civilization to the advent of AGI or SI when such an event occurs? There will surely be a rather large adjustment or it all breaks down into fiefdoms of AI Vs. no AI. Lend us your thoughts on this and we'll be happy to post them in next week's mailbox. Until then, take care, good luck and stay flexible.

Kudos to Gemini for the graphic.