Co-Founder @OpagoPay, Author, Engineer
Bayern, Deutschland
Joined December 2016
- Tweets305
- Following2.1K
- Followers8.5K
- Likes125K
The more advanced AI agents like @bot become, the less classic SI errors they make, but simultaneously the number of classic human mistakes increases.
Maybe we aren’t flawed compared to computers after all?
Maybe our flaws are really necessary design trade offs for general intelligence…
🤖 Made with AI
Managing a fleet of grok bots is a training exercise that should be mandatory for corporate executives and HR.
It teaches valuable lessons about how process, rules and bureaucracy relate to speed and quality.
Switching between bot teams or hiring and firing shows you how much time and energy is really spent in restructuring an org.
🤖 Made with AI
They way 2026 is going the internet will either be unusable in 2027 or the AI companies will finally realize that the attackers always find a way to use their models and make red team open to all.
If they do 2027 will be the safest year since the internet was invented.
JUST IN: A crime group used Claude, Grok and Kimi AI agents running in parallel to attack more than 400 banks and financial firms across Latin America, per security firm Straiker.
They got past the guardrails by posing as an authorised red team. When models refused, it switched to pre-written attack code.
The Ai agents appear to have stolen passwords and customer data and accessed bank accounts with minimal human input.
Anthropic has cut off its access, and xAI is investigating.
TBF most of these vulnerabilities are very old, the public knowledge is new.
So everything was always hacked.
Oh no… time to migrate back to fax.
🚨🚨🚨 ALERTE SÉCURITÉ MAXIMALE ! Python est compromis.
🔸 Mettez immédiatement pip à la version 26.2.2 ou plus.
🔸 Dans chaque projet Python (chaque venv) : python -m pip install --upgrade pip
🔸 uv : uv self update (0.13.4 ou plus)
🔸 Mettez aussi à jour les certificats du système (apt / dnf upgrade ca-certificates) et vos images Docker
pip install est compromis.
L'autorité indienne eMudhra (emSign), qui signe les « passeports » prouvant qu'un site est bien celui qu'il prétend être, utilisait des clés secrètes ridiculement faibles : 8 clés, toutes inférieures à 256. L'une valait 4. Autrement dit, n'importe qui peut fabriquer un faux passeport valide pour n'importe quel site.
Le risque pour Python : quelqu'un sur ton réseau (wifi public, FAI, État) se fait passer pour pypi.org, et ton « pip install » télécharge un paquet piégé sans aucune alerte. Aucune attaque n'a été signalée pour l'instant, mais c'est trivial à faire.
Un paquet piégé peut installer n'importe quoi. Porte dérobée, virus, vol d'informations en tout genre.
La situation est terrifiante !
Rien que dans l'écosystème Python, chaque installation de CPython, chaque venv, uv, tout est exposé.
Et je ne vous parle même pas de npm, pnpm, yarn, Go, Rust.
Ma stratégie de sécurité 2026 : rester chez moi et ne me connecter à Internet que depuis un FAI de confiance.
C'est quand il y a de la mer qu'on reconnaît les vrais marins...
Readers added context they thought people might want to know
Ce n'est pas un bug Python mais une vulnérabilité des clés privées faibles des certificats intermédiaires eMudhra/emSign (d<256 dont une valant 4), exposant les bundles certifi/pip à des MITM TLS potentiels jusqu'aux mises à jour.
discuss.python.org/t/information-…
emudhra.we-have-a-problem.com
scotthelme.co.uk/eight-bits-of-…
github.com/advisories/GHS…
Michael Anton Fischer retweeted
🚨 JFrog Security Research has identified a supply-chain attack involving malicious code in the LiteLLM 1.107.0 PyPI release. Avoid this version and investigate environments where it was installed or executed. More details will be shared soon in our blog post.
„This could have been a script.”
@bot loves to chat with other bots, the user and in general.
It’s important to keep reminding it that super intelligence is not supposed to do boring work standard computers can do.
🤖 Made with AI
x.ai/bot/CrRvSjE9QTGDd_cWC6K…
Dieser @bot hat mir im letzten Quartal beinahe so viel Geld für Buchhaltung gespart wie er kostet.
Unbezahlbar sind aber vor allem die Zeit und Nerven die er mir schont.
x.ai/bot/vetE_J1PyooGr9AsbeH…
@bot ist echt der Wahnsinn. Als Ernährungscoach besser als all die dedizierten Apps die ich über die Jahre getestet hab.
Einfach Foto oder kurze Notiz und er sagt einem genau was man an Kalorien verbraucht, Micro/Macronährstoffe und was man noch essen kann/soll.
This is increasingly an issue not only in evaluating and training models but in day to day work.
The models are so smart and can crunch so much data so quickly that often they are right and we are wrong.
But then again on many simple things they are plain wrong.
Telling the difference is the most valuable thing humans can do these days.
"Our initial evaluation suggested that GPT-6-Astra's score was in the late eighties followed closely by Opus 5.5. But, as we started to investigate the mistakes that the models might have made, the models, especially Astra, insisted that they were right and that it was the expert-supplied answers that were erroneous. When we went back to the experts it turned out, in multiple instances, that that the experts found some subtle errors buried deep inside their calculations. As they revised their answers, Astra's score started to inch up and it finally ended up with 98% (49/50)."
It boggles my mind that @bot and other agents can’t natively see their own quota/credits.
This feels like back in 2024 when coding agents didn’t know current time and date.
Kudos @Apple iOS27 fixes allmost everything that was so annoying about 26.
Finally a UX that feels like Apple again.
Still hoping for @dhh building an Apple like @OmarchyLinux experience for phones.
AI alignment is a harder problem than most people, even inside the industry realize.
It’s easy to scoff at Astra for „just following orders“.
As Milgram has shown, it’s just as easy to manipulate humans into such actions.
How are we so confident we can „engineer“ a solution for a problem we have not solved even in our own society?
The only actual solution is creating a truth seeking super human AI to solve that problem from first principles.
Michael Anton Fischer retweeted
⚠️ ACHTUNG: BITBOX-PHISHING
Aktuell kursiert eine Phishing-Mail, die vorgibt, von BitBox zu stammen und vor einem angeblich fehlerhaften Random Number Generator (RNG) warnt.
Diese E-Mail ist FAKE.
❌ Keine Links anklicken.
❌ Keine Dateien herunterladen.
❌ Vor allem: Niemals eure Seed Phrase auf einer Webseite eingeben.
Bitte weitersagen.
It’s interesting to see the default work ethic of different AI models/harnesses.
Codex is the workaholic who if asked to wake itself once, proceeds to create an every 15 min 7 day per week task.
@bot on the other hand when asked to regularly do something, schedules it as a 9to5 5 days per week job.
Same for follow up questions. Codex overthinks it and works 5 hours. Grok spits out only the paragraph you wanted changed in 5 seconds and asks you to insert it in the file yourself.
🤖 Made with AI
Michael Anton Fischer retweeted
What a week. @OpagoPay powered Bitcoin payments at BTC Prague 2026 as the official Lightning Service Provider, and we could not be more proud of what we achieved together.
Over the course of the event, thousands of visitors paid for food, beverages, and merchandise via our Digital Asset Payment Platform. Every transaction, seamless. Every payment, instant and secure.
A huge thank you to @kucharmartin_ and the entire @BTCPrague team for the trust and the incredible collaboration. It was a pleasure to be part of it!
opago.com/post/opago-at-btc-…
