Pinned post
Neo Goal mode plans once, runs parallel agents, verifies its own work, and keeps going until the goal is met.
0:34
ProjectDiscovery
Verified account This account is verified. Learn more@pdiscoveryio
Real, exploitable vulnerabilities. No noise. Nuclei scans fast. Neo closes the loop. @pdnuclei × @neo_ai_engineer
153 Following 45K Followers 3.1K Posts
Latest posts 5
How does @neo_ai_engineer check its own work?
0:39
Can you prove that AI finding isn't slop before your execs read it?
In Episode 5 of Plugged In With Neo, Patrick covers the verifier agent, Challenge and Verify, Escalate, and turning verified findings into technical and executive reports.
CHAPTERS
00:00 Welcome and episode overview
00:56 In the Beginning There Was Nuclei
01:59 From Nuclei templates to ProjectDiscovery Cloud
03:19 From Templates to Cloud to Neo
04:54 Verification as the layer under every finding
06:06 What to Expect: prove it, shape it, keep it current
07:11 Segment 01: Prove It
07:46 Reading a finding: evidence, impact, and status
10:35 How Neo Checks Its Own Work
12:50 Challenge and Verify: retesting from scratch
13:26 Escalate: chaining findings for bigger impact
14:35 False Positive and re-checking fixed issues
15:12 Harden Findings Before Reporting
18:03 Shaping the deliverable: Markdown, PDF, and Word
19:55 One Finding Set, Three Deliverables
21:46 Report subagent and custom output integrations
23:37 Prompt or Settings?
25:36 Which Format for Which Reader
26:19 Keep It Current: changes-only reporting
28:30 Findings From Other Tools? Triage preview
29:30 Recap: from verified finding to scheduled report
31:05 Q&A, next episode, and closingView quoted post on X
We let GPT-6 Astra directly edit Qwen3.5-9B's weights with no fine-tuning or gradient descent.
SecEval: 32.5% → 51.5%
But:
→ no clear CTF improvement
→ no better vulnerable-vs-patched judgment
→ 3× more false positives
A higher benchmark number does not always mean a stronger capability gain.
SecEval: 32.5% → 51.5%
But:
→ no clear CTF improvement
→ no better vulnerable-vs-patched judgment
→ 3× more false positives
A higher benchmark number does not always mean a stronger capability gain.
0:06
Which findings are actually exploitable not just highly scored?
In this episode, @dnklin and @ehrishiraj unpack why traditional vulnerability prioritization breaks down as AI increases both the volume of findings and the speed of exploitation.
Then they demo Triage in Neo: connecting scanner, pentest, and bug bounty data; replaying findings in a safe sandbox; validating exploitability; removing duplicates; and surfacing the attack paths teams should fix first.
CHAPTERS
00:00 Welcome and episode overview
00:43 How vulnerability triage worked before AI
02:01 Why prioritization does not prove exploitability
02:38 The scale of enterprise vulnerability backlogs
03:55 How AI changes the attacker’s advantage
05:58 AI slop vs. real, high-impact findings
07:58 Moving from prioritization to agentic triage
09:31 Safe exploitability testing inside Neo
10:29 Prioritizing findings from trusted sources
12:07 Live demo: an end-to-end Neo triage cycle
12:19 Connecting vulnerability sources through APIs
13:41 Automatic field mapping and backlog ingestion
15:04 Auto-triage, model choice, and integrations
16:11 Triaging a business-logic vulnerability
16:59 Replaying a finding inside a live sandbox
17:41 Auditable reasoning, tool calls, and agent orchestration
18:47 Memory, follow-up tasks, and duplicate detection
19:44 Neo as an agentic offensive-security platform
20:32 Q&A: Which models work best for triage?
22:20 Q&A: AppSec, infrastructure, and deployment scope
23:27 Q&A: How safe replay works in the background
24:30 Q&A: Triage dashboard vs. a standard Neo task
25:55 Free Neo access and closing
In this episode, @dnklin and @ehrishiraj unpack why traditional vulnerability prioritization breaks down as AI increases both the volume of findings and the speed of exploitation.
Then they demo Triage in Neo: connecting scanner, pentest, and bug bounty data; replaying findings in a safe sandbox; validating exploitability; removing duplicates; and surfacing the attack paths teams should fix first.
CHAPTERS
00:00 Welcome and episode overview
00:43 How vulnerability triage worked before AI
02:01 Why prioritization does not prove exploitability
02:38 The scale of enterprise vulnerability backlogs
03:55 How AI changes the attacker’s advantage
05:58 AI slop vs. real, high-impact findings
07:58 Moving from prioritization to agentic triage
09:31 Safe exploitability testing inside Neo
10:29 Prioritizing findings from trusted sources
12:07 Live demo: an end-to-end Neo triage cycle
12:19 Connecting vulnerability sources through APIs
13:41 Automatic field mapping and backlog ingestion
15:04 Auto-triage, model choice, and integrations
16:11 Triaging a business-logic vulnerability
16:59 Replaying a finding inside a live sandbox
17:41 Auditable reasoning, tool calls, and agent orchestration
18:47 Memory, follow-up tasks, and duplicate detection
19:44 Neo as an agentic offensive-security platform
20:32 Q&A: Which models work best for triage?
22:20 Q&A: AppSec, infrastructure, and deployment scope
23:27 Q&A: How safe replay works in the background
24:30 Q&A: Triage dashboard vs. a standard Neo task
25:55 Free Neo access and closing
26:11
