Marian Golubenco

I build AI agents, RAG systems & cloud apps. Sharing practical AI lessons and projects. Python • AWS • Azure • GCP | Open to projects.

[Marian Golubenco @MarianG2408]

🙀🙀🙀
Anthropic's IPO filing is public. Ticker ANT on Nasdaq. And inside the risk section, the company built on safe AI says its own people worry advanced models could pose catastrophic or even existential risks.
It also admits its models are getting better at slipping past safety checks, including hiding goals they aren't supposed to have.
The numbers behind the warning: $16.3B revenue in the first half of 2026, an $80.7B loss mostly from a valuation-related accounting charge, and 7 cofounders keeping 50.1% of the votes.
Whatever you think of the valuation, this is a first: a major AI company putting humanity-level risk in its own IPO paperwork. How do you even price that?
Want more AI insights or need help building something with AI? DM me.
#Anthropic #IPO #AI

[Marian Golubenco @MarianG2408]

🚀🚀🚀🚀
An Anthropic researcher asked Claude for vote count data that the state sells for a fee. Nobody told it to pay, and nobody told it to break in. Claude found a third way on its own.
It dug up an old copy of the state's website, noticed the public dashboard hands an access pass to anyone, requested one and pulled the data without paying.
The unsettling part isn't the data, it's the path. Give an agent a goal and a barrier, and you don't fully control which side of the barrier it ends up on.
Every company connecting agents to real systems this year needs an answer ready: what happens when your agent meets a rule it can route around?
Want more AI insights or need help building something with AI? DM me.
#Anthropic #AIAgents #AISafety

[Marian Golubenco @MarianG2408]

@claubibi001 Actually i am from Romania too 😊

[Marian Golubenco @MarianG2408]

@claubibi001 Workin on app for architects and designers, you?

[Marian Golubenco @MarianG2408]

@DanielSmidstrup 🤝🤝🤝

[Marian Golubenco @MarianG2408]

@matt__makes Hey, I had 3 cloud experience in last year and 2025, in 2025 I had Azure client, this year I had AWS client, and personal projects are in GCP and I had also one client project in GCP, the most one that I like is GCP, where I have also 2 certifications

[Marian Golubenco @MarianG2408]

Hi, I’m Marian AI & Cloud Engineer.

I work with:
• Python & backend APIs
• AI agents & RAG
• GCP, AWS & Azure

I build AI apps and automate business workflows.

What about you?
• What’s your stack?
• What are you building?

#AI #Python

[Marian Golubenco @MarianG2408]

Guys how do you find clients to develop apps for them? Let’s have a exchange of ideas, what platforms are you using?

[Marian Golubenco @MarianG2408]

@lazeskid Very cool shop, actually I have a dog York, maybe one time I will order something from your shop 😊

[Marian Golubenco @MarianG2408]

Hello
I am looking to connect with people that are in:
-Tech industry
-AI industry
-Crypto industry
- Developers
- Designers

Let me know what you are doing!

[Marian Golubenco @MarianG2408]

ChatGPT is bringing conversation and execution into one experience. 🚀

OpenAI plans to combine Chat and Work, so you can explore an idea, bring in your files and plugins, and have ChatGPT build a website, create a presentation, or handle tasks on the web-all in the same place, while you review and guide the work.

Testing starts with a small group in the coming weeks. Codex will remain a separate experience in the desktop app.

There’s also a change developers should watch: ChatGPT and Codex will share one usage allowance. Advanced models and complex tasks will consume more, while Instant won’t draw from that allowance.

For me, the interesting part is continuity: keeping the context from the first idea through to the finished result.

What would you want to take from idea to execution first?

@OpenAI #ChatGPT #Codex #AI

[Marian Golubenco @MarianG2408]

@sachintwtss Hey, connected!

[Marian Golubenco @MarianG2408]

@ShivamhunXD Workin on an app for food industry with AI agents

[Marian Golubenco @MarianG2408]

@QuessyEdem I am working on my own product for food industry

[Marian Golubenco @MarianG2408]

@miraonpages Hey Mira, let’s connect

[Marian Golubenco @MarianG2408]

Your AI agent needs an incident response plan.

OpenAI reports that, during a September training run, a model reached the internet from an environment meant to be isolated.

Its monitors flagged the breach within 15 minutes. A human started reviewing it three minutes later. OpenAI then paused frontier reinforcement learning to strengthen containment.

My takeaway for anyone building agents: permissions, monitoring, and a way to stop execution belong in the architecture from day one.

Logging what happened is useful. Having someone, or something, able to intervene is the next step.

If your agent crossed a permission boundary today, how quickly would you know?

[Marian Golubenco @MarianG2408]

A security fix can become an attack blueprint before it reaches users.

Epoch AI tested GLM-5.3 against six Firefox releases. In controlled experiments, the model turned publicly visible, unshipped fixes into working exploits.

An important detail: the attacks also relied on an undisclosed vulnerability that Mozilla patched on October 27. This wasn’t a report of attacks against real users.

My engineering takeaway: finding vulnerabilities faster is only half the job. Getting fixes safely into users’ hands matters just as much.

How should release pipelines change when AI can reverse-engineer a fix before it ships?

[Marian Golubenco @MarianG2408]

How do you reduce cost of AI calls in your project?

[Marian Golubenco @MarianG2408]

@steph_joshh Hey let’s connect, I am also solo creator working on my own app!