
building @BreachLens (BEC forensics for M365) cybersecurity nerd who ships learning in public, breaking in private python stdlib only, zero dependencies
the internet
Joined June 2024
- Tweets515
- Following43
- Followers47
- Likes274
Replying to @jk_rowling
@grok what is the Malfoid drama that all people are asking JK to address ?
2
155
cogio retweeted
Since we all had a fuck you money week, I’ll reintroduce you to the three most life changing purchases I’ve ever made.
Thank me later. No ref links just straight alpha.
Replying to @merill
good timing — most of my week is AI against M365 audit logs. biggest lesson so far: it's great at flagging weird inbox rules and OAuth grants, but you have to feed it the UAL in small well-scoped slices or it drowns. curious where you two landed on guardrails for admin-side AI
1
26
Replying to @tdinh_me
congrats man. the ASO grind is real — i once spent two weeks on keywords and got outranked by an app that just put 'free' in the title. honestly the tiktok dance will outperform all of it, post the rehearsals
223
Replying to @cyb3rops
this matches what i've learned building on LLMs the hard way: every static filter i shipped got walked around within a week by a prompt i never imagined. what actually held up: shrinking what the model can *do* — least-privilege tools, read-only defaults — and logging everything
218
Replying to @marclou
principle 1 stings. spent months polishing a free tier nobody converted from before realizing free users were costing me support hours and feature noise. the 'leeches' framing is harsh but the support inbox doesn't lie. curious which of the 32 you'd say founders break most
1
125
Replying to @brutecat
the part that gets me is how much of this is patient methodology with AI doing the tedious 90% — recon, enumeration, fuzzing the boring corners — and human intuition landing the actual bug. industrialized patience as a $500k skill. going into my weekend reading, congrats
302
Replying to @cyb3rops
matches my experience building with LLMs for security work. shipped a static guardrail layer once, watched it fold to the third creative prompt. treating it like detection engineering — assume bypass, log everything, keep iterating — is the only framing that survived real users
353
Replying to @ClaudeDevs
Pretty meta — vibe-coding BreachLens (BEC forensics for M365) with Claude Code, so having it lint my own security code is welcome. The catch I'd want most: accidentally logging raw email headers/tokens. Most common own-goal I see in early-stage security tools.
1,473
Replying to @TheDFIRReport
The on-prem dsa.msc move has a tidy Entra cousin: actors register service principals with 'svc-admin' or 'Sync-' prefixes that blend into tenant inventory. Same play — name camouflage. Pair AD 4720 hunting with Entra app registration auditing.
40
Replying to @haveibeenpwned
Employer info is what tips this from credential-stuffing fuel into BEC ammo. The follow-up phishing pretexts I see in M365 audit logs almost always reference role/department detail straight out of dumps like this. Time-to-weaponization is days, not months.
39
Replying to @TheHackersNews
the persistence is the easy story. these increasingly pivot into M365 same-session — inbox rules + forwarders before anyone clocks the second C2. malware-only IR misses where the real damage lives. exactly the gap i built breachlens.io for
253
Replying to @DFIRTraining
the wild part isn't the channel existing — it's 9k people deciding that trade was worth it. one cross-exam question and the entire report's credibility is done. weird hill to die on for a few bucks saved on a license.
1
104
Replying to @HackRead
calendar invites slip past the 'hover before you click' instinct — they land in your calendar so trust is implicit. for token theft the real signal isn't the invite, it's the sign-in from a weird ASN then an inbox rule ~90s later.
13
Replying to @Mandiant
vishing → AiTM → MFA bypass is the one i hate triaging. by the time helpdesk closes the ticket, inbox rules + token replay are already in motion. when i pull UAL the whole compromise fits in <15 min. exact reason i started breachlens.io
68
matches what i hit building agentic tools against Graph — Opus already knows ~80% of the surface from training, so the skill mostly just needs scoping + pagination/throttle handling. code mode + a thin cli has been way more reliable than json tool-call for batch graph ops
2
61
Replying to @ipurple
the polished panel UX is the tier shift — one-click grant → token → rule → exfil collapses the IR window to whatever UAL latency is. why i built breachlens.io: stitch those four signals into one timeline so you stop losing minutes pivoting consoles
18
Replying to @EsGeeks
the 'AV says clean → not clean' framing is half the DFIR mindset. Get-WinEvent + Get-MpThreatDetection + Get-ScheduledTask is on my BEC IR cheatsheet — by the time defender goes red the evidence is usually overwritten. solid thread.
51
Replying to @Rendani666
three departments, three 'separate' tickets — that pattern is exactly why i started breachlens.io. each team only logs their slice (IT: password reset, finance: mismatch, payroll: callback) and the actor lives in the gap between the consoles. saved this one.
15



