information security - bug bounty - motorsport - karting - memes - shitposts
Kaurna Country
Joined December 2007
- Tweets14.6K
- Following1.1K
- Followers6K
- Likes38.4K
Let's be honest, those of us with agentic tooling aren't handcoding it - are we.
Have you reviewed the actual prompts recently?
I reviewed mine and discovered my prompts had ballooned!
Review the prompts, and see what could be a thin skill & tool call instead!
I cut 25%!
1
7
409
I love it!
It's like directing a smart and motivated junior to mid developer to implement what I plan and design.
I get to do the fun bit now without getting bored while writing the code. It's finished that feature by the time I come up with the next one!
programmers who spent years learning to code before ai
how does it feel watching a beginner build in minutes
9
993
I need to test more, but my experience so far matches this.
Never seen a refusal on Deepseek at all. Ever.
Abliterated models are a hoax. 3 debunked fake claims:
1) attackers don’t use them
2) they don’t actually find more/better vulnerabilities
3) they don’t lead to fewer false positives
I’m extremely skeptical of “security” products and companies offering this as a service.
Initially I was really excited, but after running them at scale and investigating incidents they perform worse that the vanilla model.
omg hit tweet
to learn about what the quoted post was actually referring to, check out the doco at docs.docker.com/ai/sandboxes…
but, always remember, docker is not a sandbox (on its own).
4
23
4,741
.@deepseek_ai I've been reviewing my logs & spend over the recent weeks, it looks like cached tokens aren't being charged at the advertised rate - they're charged at the regular rate. Cost is 6x!
Is there an issue your side, or has the price for cached tokens changed recently?
Seriously, harness and tool calls folks, this isn't 2025 anymore 😂
Hacking is all AI now but the fundamentals are still yielding bugs, like reverse DNS lookups.
The difference is that my agents use the tools now. I just point the agents at my custom tools and tell it to turn the README's into skills.
This is hakrevdns. It's very simple, it does reverse DNS lookups as fast as possible. It's a great way of discovering domains and subdomains owned by a company when you know their IP address range(s).
github.com/hakluke/hakrevdns
57
5,426
"Select daybreak red customers"
I.e. Unless your company is global or a household name in infosec, you don't get access.
Seems to be how these are all gated these days.
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.
GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.
openai.com/index/introducing…
You know what?
I like surprises.
So I've decided not to worry about how long triage and payment takes, I'll just keep hacking & keep submitting.
Then when something happens on a platform, it'll be a fun little surprise!
Maybe it'll be a dupe, maybe some money? Who knows!🤣
1
8
647
Oh btw I got access to @OrcaRouter cyber tier last night.
Had some agentic tests running already I didn't want to interrupt to test it, will run some tonight with an abliterated model and see how it goes!
Very close to giving up on @AnthropicAI CVP RT. Still pending 🤦♂️🤦♂️
3
17
1,696