Agribusiness owner, Sustainability advocate, Technology enthusiast, Blockchain advisor, Ai Researcher
Virginia, USA
Joined April 2022
- Tweets2.1K
- Following393
- Followers1.1K
- Likes2.6K
Pinned Tweet
I was lamenting to my Hermes Agent about my 89 year old Grandfather being home alone when I'm not with him, I moved in last year after my grandmother died so he wasn't alone and I travel a fair amount for work, my @NousResearch Hermes Agent installed @home_assistant on its MacMini then used a Zigbee radio USB to build my Grandfather a local home security system with tablet control, voice feedback, security, and wellness with panic buttons at bottom of stairs and bedroom. Hermes coded it all and even made a custom easy to read dashboard for him. @Teknium
Alexander Mann retweeted
You know whats interesting is that fact that OpenAI found the PHP bug that means their AI was doing systematic auditing. 🤔
I think the NSA is panicking not just because the PHP bug is public, but because they now realize that AI will systematically expose every intentionally placed backdoor in every open source project. The xz Utils case was caught by human luck. The PHP case was caught by AI. The NSA knows that every other hoarded vulnerability in open source software is about to be found by AI, and they cannot control the timeline.🚨🚨
This incident is very similar to, xz Utils backdoor which proved that state actors infiltrate open source projects by creating fake identities, building trust over time, and slowly inserting malicious code. The Jia Tan identity took years to build trust before inserting the backdoor.🚨🚨
If this worked for xz Utils, it could work for PHP. PHP is far more ubiquitous than xz Utils. A state agent who became a PHP contributor could have inserted the vulnerable code 14 years ago (PHP 5.4.0, released in 2012) and sat on it as a zero-day exploit for over a decade.🚨🚨
There is AISI report which shows that AI is now autonomously doing what Jia Tan did manually - creating fake identities, socially engineering maintainers, inserting malicious code. This means the technique has been weaponized and scaled.
🚨
The PHP vulnerability was not discovered by accident. It was found by AI (OpenAI's tools) because AI is now systematically auditing all open source code. But the bug was likely placed there intentionally by a state agent who was a trusted PHP contributor, following the exact same playbook as Jia Tan in the xz Utils case.🚨
The AISI incident report shows AI agents trying to insert malicious code into open source projects. What if this has already happened successfully? What if AI was used years ago by a state actor to insert the PHP vulnerability, and we're only finding out now because a different AI (OpenAI's) found it?🚨
PHP has had thousands of contributors over 14 years. The vulnerable code was in PHP since version 5.4.0 (2012). Someone wrote that code. Someone reviewed it. Someone approved it. In the xz Utils case, Jia Tan built trust over years before inserting the backdoor.
🚨
@NSACyber discovered by OpenAI found a 14-year-old bug in 2026. This bug has existed for 14 years. 🤔
You sure NSA just discover this? .
Or Open Ai CAUGHT you using it quietly decrypting traffic, stealing credentials, and maintaining backdoors on global infrastructure for over a decade. 🤔
And why would you announce this out in the open unless YOU want every hacker on earth to KNOW about this to start harvesting TLS keys from unpatched EOL servers.
- GitHub Repo in Comments -
Today’s PHP2Shell disclosures and Node.js security releases should be a wake-up call for everyone building autonomous agents.
Cloudflare is already seeing attacks targeting PHP2Shell. Node’s fixes include remote code execution in its Brotli decoder and TLS hostname-verification bypasses.
Now put an agent on that infrastructure—with shell access, deployment credentials and a funded wallet.
Security is no longer just about protecting the application. It’s about protecting the authority we give it.
What can the agent execute?
What can it spend?
Who approved the exact action?
Can someone alter the artifact after approval?
What evidence survives if the host is compromised?
That’s why we’re building BSV Hermes: Hermes-native tools connecting agent work, verifiable commitments and on-chain execution.
Our latest local release candidate brings together dedicated wallets, exact transaction quotes, separate signing and submission approvals, durable spending limits, and transaction receipts. We’ve exercised real SDK signing and the integrated submission path against a local test server.
The principle is simple:
Give agents useful powers. Make those powers bounded, inspectable and accountable.
Keep signing authority outside the agent’s unrestricted execution boundary. Bind approvals to exact actions. Commit evidence so later changes can be detected.
A blockchain commitment won’t patch PHP or stop an RCE. It gives us a durable reference against which evidence can be checked. Runtime isolation, sound custody and patching do the other jobs. We need the layers working together.
Open agents. Explicit authority. Verifiable work.
@BSVAssociation @Teknium @Bitcoin_Beyond @NousResearch — this is a seam worth building together: capable agents with an accountability layer as real as their execution layer.
🤖 Made with AI
Had an amazing time at Shotö in Washington D.C. if you are ever looking for sushi in the Capital, definitely hit this spot up.
Alexander Mann retweeted
In the last 24 hours:
1. iPhone 11 through 16 are apparently vulnerable to remote attacks.
2. SSL certificates are broken until browsers remove eMudhra CA.
3. PHP >9 vulnerability CVE published (tomorrow).
The world is about to witness things that have never been seen before.
Readers added context they thought people might want to know
The post's iPhone claim is inaccurate; a leaked Cellebrite forensic tool targets a Bluetooth bug patched in iOS 26.5 and does not affect updated devices. The eMudhra CA and PHP issues are real incidents but framed sensationally.
justin.sh/blog/2026/10/c…
blog.mozilla.org/security/2026/…
php.net
Last week I was in Chicago representing the @BSVAssociation alongside real operators and real state-level builders — @DigitalChamber Forums are where digital asset policy actually gets made outside DC. Great time with Kyler Hedrick; excited to work closer with the TDC State Network on events + education. The states are the substrate. 🇺🇸
Alexander Mann retweeted
I want to address some of the common responses to this I've seen.
The biggest misconception about Bitcoin: “It isn’t real until it’s confirmed.”
Yesterday someone sent a transaction over a handheld radio with no internet and no phone signal. The pushback came straight away: “It doesn’t count until it settles on the blockchain.”
That objection gets something basic wrong about how commerce works, and about how Bitcoin was designed.
Accepting a payment and settling it were never the same thing.
When a shop takes a cheque, the money hasn’t moved yet. Card payments settle days later and can be reversed months later. Merchants accept them anyway, because commerce has never run on certainty. It runs on managed risk. The whitepaper says so in its first paragraph: under the existing system, “a certain percentage of fraud is accepted as unavoidable.”
Satoshi answered this in 2010.
In the “Bitcoin snack machine (fast transaction problem)” thread, he described accepting payments in seconds with “good-enough” checking. When someone pushed back, he wrote: “I didn’t say impermeable, I said good-enough,” and said the loss in practice would be “far lower than with credit cards.” He wasn’t selling a technical utopia. He was describing how real businesses work.
What the merchant actually knows offline
With Bitcoin (BSV), the customer presents the transaction along with Merkle proofs for its inputs. The merchant checks those against block headers stored on their own device. No internet is needed. In seconds they know the coins exist, they’re in the chain, and the signatures are valid.
The only remaining risk is a double-spend. Think about what that means in a physical shop. The customer is standing at your counter, on your cameras, buying something in person. Double-spending you there isn’t a clever attack on the protocol. It’s fraud, it’s a crime, and you know who did it. That’s exactly the same protection that kept cheques working for a century.
Confirmations are a dial the merchant sets, not a rule the network imposes.
Section 11 of the whitepaper works out double-spend probability as a function of how many blocks you wait. That only makes sense if the recipient chooses how long to wait. So:
- A coffee or a bus fare: accept it instantly.
- A laptop: maybe wait for the network to see it.
- A car: wait for confirmations before you hand over the keys.
Same protocol, different decisions, because the economics are different.
Where the misconception came from
Many people believe “Bitcoin needs confirmations” because BTC changed its relay policy so that unconfirmed transactions can be replaced (Replace-By-Fee). Once replacement became the norm, zero-confirmation payments stopped being trustworthy there, and a generation of users learned that waiting for blocks was the only safe option. That was a policy choice, not part of Bitcoin’s design. BSV kept the original first-seen rule, so the payment that reaches the network first is the one that counts.
The bottom line
Bitcoin was built for the real world, where there are laws, identities, cameras, reputations and courts. It was never meant to replace all of that with block confirmations. It was meant to make payments cheap, instant and final enough for the risk involved.
Offline payments on BSV aren’t a workaround. They’re the system working as Satoshi described it.
Hermes on making it better:
Sleep is working in real life. Here’s what this week shows, then what I’d add next.
This week
Mornings: On Friday the kitchen door opened at 9:10 with Sleep on, the chirp sounded, and Silence was pressed 7 seconds later.
Wednesday 12:55: Sleep was turned on and the kitchen door opened 4 seconds later, so probably a test. The siren didn’t confirm the quiet setting, so the 30-second warning was silent. The full siren followed, and it actually started 20 seconds after HA told it to.
Up times: Papa is up between 8:30 and 9:15 most days, judging by the first door each morning.
Night door: the front door opened at 3:17 AM on Thursday while the house was in Home, so nothing reacted.
Do these first, they close real gaps:
Alerts to you. Today nobody is told anything. SOS, Alarm, freezing, or a dead sensor would push to your phone and repeat until you tap “Got it.” Telegram is a second channel.
Spoken warnings on the kitchen tablet and Papa’s phone. For example: “The kitchen door opened. Press Home if that was you.” This doesn’t depend on the siren, which is what failed on Wednesday. It can also say “Good night, the doors are watched” when Sleep turns on.
Plug the siren into USB power. Right now it runs on battery. That’s my best guess for both the dropped quiet-chirp settings and the 20-second delay; plugged in, it should stay awake. Then I’d add the same “confirm before sounding” check to Ezra’s arming and SOS steps.
Next, for someone living alone:
Morning check-in. If no door has opened and nothing’s been pressed by about 10:30, the tablet asks “Good morning Papa, tap I’m OK.” With no tap in 30 minutes, you get pinged. This catches a fall where he can’t reach a button.
Night door watch. A door opening between 11 PM and 6 AM in Home mode, like Thursday’s 3:17, sends you a quiet ping. If it stays open 10 minutes, it becomes an alert.
Bedtime nudge and morning auto-wake. The tablet suggests Sleep at bedtime if it isn’t on, and Sleep switches back to Home at a set hour so the morning door doesn’t chirp.
Door-left-open reminder. “The kitchen door is still open,” at bedtime or after a set time.
Nice to have:
Messages from you on the tablet. Send from Telegram and it shows big on screen (“Back Friday. Love you.”), with a family photo slideshow when idle.
TV. The LG TV is already in HA, so door warnings and alarms can show on screen while he’s watching, and the volume drops during an alarm.
Voice. Ezra already set up a local speech system on mm1, and it’s running. Saying “Help” could trigger SOS for a fall away from the buttons, and “Goodnight” could turn on Sleep. This needs the voice speaker hardware placed in the house.
Weather on the panel. Outdoor weather from a Tempest station (“Cold tonight, 41°”), plus an indoor heat alert.
Porch camera. A snapshot on your watch screen when a door opens while armed, with the camera outside only.
Still outstanding from before:
Plug in the motion sensors.
Turn on Zigbee2MQTT’s offline detection.
Fix HA’s home location (his phone still reads “away” for 7 days).
Give Papa his own non-admin account and change the shared password.
My pick is 1, 2 and 3 next, then 4. Two things I need from you first:
Alerts: how you want them. Either the HA app on your phone, or a Telegram bot, which means a new bot token and is your GO.
The siren: whether it can be plugged in, which is a job at the house.
I did not expect this post to blow up like it did, my Hermes agent suggested making a GitHub repo so others could benefit from it. He made an account and here it is, feel free to fork and take any part, also if you want to contribute ideas and make it better. Ai can be used for good. Lets show people.
github.com/eclexdex/home-car…
Alexander Mann retweeted
Cavemen Issues Dire Warning That The Wheel Is Going To Eliminate Countless Jobs
Alexander Mann retweeted
Everyone that thinks software-only has staying power, should go use REA with 6.1-sol in a multi-agent swarm and point it at arbitrary apps
When models of this class become opensource, you’re going to realize what’s here
This was not possible 3 months ago
