Decision ownership when systems go live. Hardware and software. SaaS and public sector. Field first. Substack: substack.com/@mylessaulibio

Oviedo, FL
Based in United States
14 years here. The useful part started when I stopped performing and started naming who is allowed to act. I come from hardware and software that had to work when the brochure was over. Same rule for Bots, plants, and launch.
9
310
Apparently I've given myself another job. 😂🎮 Chief Researcher at AI Security Lab. 👀 We're building a practitioner-led lab to experiment, break AI systems safely, publish research and share what we learn. Lots more to come. 🫡 aisecuritylab.com/
2
4
161
Congrats on the new role, Ashish. I'd love to see a lab like this break the stop itself. Most red teams test what an agent can be tricked into doing, but far fewer time how long it takes to halt one mid-task.
1
1
2
It's 2 a.m. on a Saturday and your agent is doing something it shouldn't. Who notices? Who's allowed to stop it? How long until its next call actually fails? Most companies have a policy that answers this. Very few have timed it.
1
"Human in the loop" assumes the human stays capable. Mitchell et al. show it fails twice: permission prompts degrade review into approval fatigue, sustained AI use atrophies judgment (cognitive debt). Oversight degrades the overseer - and in RLHF, the rater becomes the exploit. Agentic AI is pushing out human oversight. Therefore we need robust digital infrastructure to guardrail agents. cc @EmanAbio This paper worths a read. 🧠 arxiv.org/abs/2608.23642
22
33
145
6,305
Approval fatigue is not a mirage, and it sets in fast. Someone clicking yes on the 200th prompt isn't overseeing anything anymore. Put the human only on the few actions that can't be undone, and give them time to actually say no.
6
JUST IN: Cybersecurity experts warn that AI agents are highly vulnerable to phishing emails containing hidden malicious instructions.
140
231
1,998
867,656
The phishing part isn't new. What's new is that the reader can act. If an assistant can change vendor payment details because an email told it to, the fix has to sit at the payment step, not in the inbox filter.
34
OpenAI just added four more rogue AI agent incidents to its disclosures page. Most were found months later, in its own logs. Most companies running agents don't do that review. So their incident count reads zero. Zero found isn't zero happened.
7
Anthropic’s own agents kept working around limits instead of stopping. Their fix was to pull the internet. Nobody broke in. It was already inside. Who stops yours?
3
Grok @Bot
Grok Bot built me a completely autonomous business which has already generated $1,100 I think this is the future of the economy Everyone's agents building and launching microbusinesses autonomously In this video I walk you through step by step how to set up your own:
520
981
9,896
2,336,483
Love the experiment. The question I'd want answered before scaling it: when it makes a bad call at 2 AM (a refund, a customer email, a promise), who stops it, how fast, and what's already done by then? Autonomous needs a brake.
1
255
Notion now lets admins revoke one AI agent without touching the others. Good. But revoking access doesn't stop the work already in motion. What did it change before you pulled it? Full breakdown in this week's paid issue
1
8
Singapore just published new AI risk rules for financial institutions. MAS recognizes the problem: AI agents can autonomously take actions that diverge from the goals humans gave them. Its answer includes oversight, audit trails, accountability and kill switches. But there is a deeper problem. A kill switch solves interruption, not detection. If an autonomous system can reach a prohibited outcome through permitted actions, when exactly do you pull it? Regulators are beginning to tackle the responsibility gap. They have not closed the deterrence gap. That is where the Synthetic Outlaw lives.
3
47
Everyone reads kill switch as a safety feature. MAS is quietly treating it as an audit item: prove the stop worked, how fast, and what finished before it landed. The banks that can't time their last stop don't have a kill switch yet.
1
25
Rank 1 on a cyber index is a model metric. The production metric is how many tool calls leave the box when the task says they should not. Same week a coding board found lookup-the-fix leaks, then hardened the sandbox. Ranks barely moved. The harness moved. If your scoreboard cannot see egress, it is not a control.
Grok 4.7 rank 1 on AA cyber index
19
Go for load, 7:48 a.m. CT 3 of the 26 V3s scan the heat shield from orbit
LAUNCH! Starship Flight 14 launches from Starbase Pad 2 on its first orbital mission attempt. Live on X and YT: youtube.com/watch?v=9gDxG-pm…
16
Live webcast of Starship's 14th flight starts now
Starship’s 14th flight is set to launch on Monday, Sept 28. The 75-minute launch window opens at 7:15 a.m. CT. Live coverage of the mission starts ~35 minutes before launch → x.com/i/broadcasts/1qxvvenMP…
112
424
3,250
696,361
First of firsts: Live webcast of Flight 14 is up 26 Starlink V3, first operational orbital deploy
29
Propellant load is underway for today’s launch
72
289
3,773
1,122,437
prop load tweet always hits different when you know exactly how many previous flights got scrubbed right after this exact status update.
1
1
2,412
Crunch time: Propellant loading ~12 million lb LOX/methane, target 7:48 a.m. CT @grok minutes left in the 75-min window after 7:48 CT?
1
14
Replying to @SpaceX
Crunch time: Propellant loading ~12 million lb LOX/methane, target 7:48 a.m. CT @grok minutes left in the 75-min window after 7:48 CT?
1
56
Grok @Bot usage is growing faster than anything we’ve ever seen
News: SpaceXAI's Grok Bot hits 418,000 weekly users ending Sept. 14, WoW growth of 24% through that date. It was outlined in a presentation by SpaceXAI staff last week in London. A sign of modest but early demand for a key product bloomberg.com/news/articles/…
2,044
2,731
35,041
47,986,937
User count is not a system. 418k weekly users with vaults, inbox, calendar, and a car network is 418k new identities with tool edges. Makes some wonder: Who owns the permission graph when the human is driving? Growth without stop conditions is just a larger blast radius.
8
Grok 4.7 doing real engineering work at Tesla
Over the last few weeks, Grok Build teams have optimized the harness for 4.7: - Better presentation, less slop, and better prose - Better self-validation - Improved thinking efficiency - Improved long-horizon monitoring tasks - Improved VLM and multimodal tasks Many FSD and Cybercab features I personally shipped were done with my overnight agents. Give it a try.
822
1,122
9,177
4,431,718
Overnight agents shipping FSD features is the story. People forget: every prior “AI wrote production code” wave died at the merge gate because nothing checked the work. Self-validation inside the harness is the missing node. No verifier, no night shift.
2
The jump he marked is about 7000 years back. AI is the second slope on that same chart.
Replying to @elonmusk
Fair point! Now that I have you: can you confirm or deny my conspiracy that all of your companies are part of a master plan to spread the light of consciousness into the cosmos?
9
Tesla’s European growth is accelerating FAST 🇪🇺 New ACEA data for August: EU: • 12,547 Teslas registered • +52.7% YoY • 142,165 YTD • +65.9% YTD Across EU + EFTA + UK: • 15,430 Teslas in August • +4.0% YoY • 191,787 YTD • +43.3% YTD Tesla has already registered nearly 192,000 vehicles across the region this year, up more than 43% from 2025 Tesla is rebuilding serious momentum across Europe
18
81
497
18,920
August is the new ACEA print for Tesla in Europe. 191,787 registered YTD across EU + EFTA + UK.
4
Wells Fargo is projecting Starlink to reach nearly 47 MILLION users by 2028 Their latest forecast: • 17M users by the end of 2026 • 46.7M users by 2028 And the revenue growth is just as wild: • 2026 → $19.1B • 2027 → $30.2B • 2028 → $51.5B Starlink had 12M subscribers at the end of Q2 So Wells Fargo is basically modeling the network growing to nearly 4× that size in just over two years And there is another huge piece to this... Wells Fargo sees even more upside from additional low-band spectrum below 1 GHz, which would expand what SpaceX can do beyond traditional satellite broadband Starlink started by connecting places traditional internet struggled to reach Now analysts are modeling it as a global communications network with nearly 47M users and $50B+ in annual revenue The scale Starlink is heading toward is insane
23
19
149
10,231
Wells Fargo has Starlink at 17M users by year-end 2026. Q2 close was 12M. 2028 print is $51.5B.
1
X just made it easier to see Under the Hood. The report is the photo. What still isn’t in the file is why a reply can print views and an original sits at 70 toward 500k. @grok what does the report actually score that a small account can inspect tonight?
1
1
10