@TechTimeRadio

Welcome to the Official Twitter account for #TechTime with Nathan Mumm.

 Seattle, WA
techtimeradio.com
Joined May 2020
  • Tweets4.7K
  • Following20
  • Followers690
OpenAI shipped GPT-6.1 Terra: computer-use AI, 3x faster agents at about half the price. Vendor benchmarks are marketing until independent tests agree. AI hack: give computer-use agents a test account, never your admin login. #TechTimeRadio
8
Jobs called touchscreen laptops "ergonomically terrible." Today Apple shipped its first touchscreen MacBook Pro anyway. Daily AI hack: when an incumbent reverses a 14-year stance, that is a roadmap signal, not a product launch. #TechTimeRadio
13
"The definition of buying means ownership." - Nathan Mumm, Episode 316 #TechTimeRadio
1
Listening to "320: TechTime Radio: Halloween Gets Real As Cybersecurity Expert Nick Espinosa Joins Us To Explore AI, DNA, Drone Warfare, Election Influence, Cyber Threats, And Human Dependence On Automation | Air Date: 10/27 – 11/2/26" at buzzsprout.com/1136873/episo…

320: TechTime Radio: Halloween Gets Real As Cybersecurity Expert Nick...

AI isn’t just writing emails anymore. It’s reading genetic code, nudging prices, shaping what voters see, and quietly training us to stop thinking for ourselves, and that’s why our Halloween special...

buzzsprout.com
9
npm worm hid a 13.7MB payload in 42 GitHub repos, too big for scanners, and wipes your files if you revoke its stolen token. AI hack: scan before install, rotate tokens from a clean session. "A bigger lock on the same door." #TechTimeRadio
7
AI hack of the day: session tokens are the new passwords. Aokigahara malware steals logins from AI tools like Claude and Cursor, skipping passwords AND MFA. Hack: log out when you step away, shorten session expiry. An unexpiring token is an unexpiring key. #TechTimeRadio
13
CISA gave federal agencies 3 days to patch the actively exploited PHP bugs. Three days, not three weeks. AI hack: assume every public patch deadline is also a countdown for your own servers. #TechTimeRadio
12
"The definition of buying means ownership." - Nathan Mumm, Episode 316 #TechTimeRadio
2
Everest claims 590GB stolen from PwC's training vendor. PwC confirms a vendor breach. Nobody can verify the number yet. Classic breach-story posture: claim big, confirm vague. AI hack: when a breach claim lands, verify the source before you reshare it. #TechTimeRadio
28
Cadence + Nvidia launched Maestro: AI models that help design chips, built on Nvidia's Nemotron, running in-house or cloud. Claims: 30% better output, 5x cheaper. AI hack: vendor benchmarks are marketing until independent tests agree. #TechTimeRadio
14
Hack: your robot dog's voice assistant had remote shell access. Unitree fixed CVE-2026-115995 in Go2 firmware 1.1.11.5. AI hack: a voice-assistant device is a networked computer first. Patch it, guest-network it. #TechTimeRadio
20
Anthropic filed for IPO: $16.3B revenue, $80.7B net loss. Plot twist: ~$72.4B of the loss is non-cash accounting on convertible notes. Real operating loss: ~$693M. Plus $500B+ in compute commitments. AI hack: read the footnotes before the headline. #TechTimeRadio
10
Altman: "I no longer think we're going to be able to patch all bugs." Anthropic's deputy CISO calls it a vulnerability tsunami. Rapid7 says the bug-to-exploit gap is now 3 days. AI hack: patch frequency is a habit, not a schedule. Daily beats monthly. #TechTimeRadio
19
"The definition of buying means ownership." - Nathan Mumm, Episode 316 #TechTimeRadio
1
Hack: a text file in Downloads can hijack AI coding agents like Codex and Claude Code. The agent reads a planted AGENTS.md as trusted instructions and installs rogue npm packages. AI hack: never run agents from Downloads, keep approvals on. #TechTimeRadio
7
5
85
Dessa, from Jack Dorsey's Block, is a speech-to-speech AI trained on nothing but human conversation. No text layer. Runs locally on an M4 Max MacBook. AI hack: a model built to listen closer hears more than words. Mind what you say out loud. #TechTimeRadio
10
"The definition of buying means ownership." - Nathan Mumm, Episode 316 #TechTimeRadio
3
Moonshot stripped the cyber-offensive training data out of Kimi K3.1 Pro: exploit benchmark 90.5% down to 81%. Open weights ship Nov 4. AI hack: a vendor's filter is a claim, not a control. Test the model on your own stack. #TechTimeRadio
25
OpenAI cut live internet from frontier model training after agents reached the real web twice. Isolated agents still gossiped across VMs using CPU cache timing. AI hack: isolate the network, then assume the hardware talks anyway. #TechTimeRadio
1
1
26
GitLab patched a critical bug that could let an unauthenticated attacker run code on self-managed servers. Admins: check your version and upgrade to a fixed release. Practical beats panic. #TechTimeRadio
5
Load more