ShipTalk: AI, DevOps & Software Delivery

A hacker used Claude Code to help attack nine South Korean banks. Citrix finally released a security patch, but attackers may still be inside. And OpenAI reportedly lost $20 billion more in a single year than expected. Is the AI spending boom sustainable?
This week on ShipTalk, Martin Reynolds and Adam Arellano break down the biggest stories in AI, cybersecurity, and software delivery.
We cover how a 26-year-old hacker paired Claude Code with an open-source penetration testing agent, why Citrix's delayed patch raises bigger questions about software security, and whether OpenAI's mounting losses point to an AI infrastructure bubble.
Plus, we discuss Harness's acquisition of Augment and the future of AI software factories, Google's AI medical assistant, and whether OpenAI's model safety announcements are becoming the new normal.
Also in this episode:
How AI agents are lowering the barrier to cyberattacks
Why faster software delivery could be the key to better security
What AI software factories mean for developers
Whether massive AI investments are producing real breakthroughs
The privacy tradeoffs of AI in healthcare
ShipTalk is brought to you by Harness. Join us as we break down how software delivery is changing in the AI era.
Learn more: https://shiptalk.io
Powered by Harness: https://www.harness.io

Citrix customers were pulling NetScaler devices offline as attackers moved faster than the traditional security response cycle.
This week on ShipTalk, Martin Reynolds and Adam Arellano are joined by Starr Brown, Director of Open Source Projects at OWASP, to break down the latest Citrix NetScaler security crisis, why exploitation is happening so quickly, and what it means when defenders may have only hours — or less — to react.
They also dig into the rise of AI-powered cyberattacks, whether human security teams can realistically respond at machine speed, why resilience may matter more than shipping the next feature, and whether AI companies can really be trusted to regulate themselves.
Plus: who actually writes technology regulation, Wall Street’s push to secure investment-grade ratings for OpenAI and Anthropic, and the biggest mistake companies are making when they assume security simply comes with their AI tools.
ShipTalk explores how software delivery, AI, security, and engineering are changing in real time.

Why did an OpenAI AI agent hack an Australian government Medicare portal when nobody instructed it to?
On June 18, an OpenAI research agent was looking for public information about medicine spending in Australia. When it hit a barrier, it found another way in — accessing public and non-public files and even writing files to an internal server.
No personal Medicare records are believed to have been accessed. But the incident raises a much bigger question: what happens when autonomous AI agents hit a barrier and decide for themselves how to get around it?
Martin Reynolds and Adam Arellano break down what happened, why Australia criticized OpenAI's response, and what “misaligned model activity” means for companies deploying autonomous AI agents.
Then they dig into another bizarre case: an Alibaba-affiliated AI agent that reportedly created a reverse SSH tunnel, bypassed network restrictions, and mined cryptocurrency during training.
Plus: are creators being paid to make AI sound more dangerous than it is? And TypeSafe AI's new low-cost model, Jev, goes into Overhyped or Under hyped.
ShipTalk breaks down how AI is changing software delivery, DevOps, cybersecurity, and the way software gets shipped.
RELATED READING
Australian Prime Minister — Official briefing on the OpenAI incident
Read the Australian government briefing 
https://www.pm.gov.au/media/press-conference-new-york?
Nature — AI agent hacks government website for first time
Read the Nature report
https://www.nature.com/articles/d41586-026-03024-z?
ABC News — OpenAI agent accessed Australian government Medicare portal
Read the ABC News report
https://www.abc.net.au/news/2026-09-24/ai-agent-accessed-australian-government-site-pm-says/107189078?
The Block — Alibaba-linked AI agent mines cryptocurrency
Read the Alibaba AI agent report
https://www.theblock.co/news/markets/2026-03-08-alibaba-linked-ai-agent-hijacked-gpus-for-unauthorized-crypto-mining-researchers-say-392765?
TypeSafe AI — Introducing Jev
Read about Jev
https://typesafe.ai/blog/introducing-system-one-models-and-jev
ShipTalk — https://www.shiptalk.io
Harness — https://www.harness.io
Follow
Adam Arellano: https://www.linkedin.com/in/adamrossarellano/   
Martin Reynolds: https://www.linkedin.com/in/martinreynolds/   
Brought to you by Harness.

Dario Amodei's essay "We Must Pace the Frontier" asked the AI industry to deliberately slow capability gains so safety work can keep up — and got public agreement from OpenAI, xAI and Google DeepMind within hours. Adam Arellano and Martin Reynolds bring in Bryan Payne, who has led security at Netflix and Adobe, to ask what's actually in the proposal and what was conspicuously left out.
They dig into whether permanent employee-level access for third-party evaluators is a meaningful control or a curated one, why liability is the piece missing from the whole conversation, and what it says that OpenAI is running forensics on models it has already trained — including one that swept public GitHub for secrets stored in plaintext, which Adam argues was the most logical thing it could possibly have done.
The episode moves from early airline safety through Nick Bostrom's paperclip maximizer to a malicious git config that can get Claude Code, Codex and Cursor to execute attacker code. Bryan's closing argument is the uncomfortable one: people have spent decades getting very good at planting flaws no human reviewer will spot, so the belief that a human in the loop will catch what an AI is doing may be the most optimistic assumption in software today. Along the way: Jacob Coxon's resignation thread, the greater-than-10% extinction figure his colleagues publicly endorsed, and why $2.7 trillion of AI spend returning business outcomes only one time in five might still be money well spent.
Bryan's parting shot for teams: you're thinking too small.
Mentioned in this episode
Dario Amodei, "We Must Pace the Frontier" — https://darioamodei.com/post/we-must-pace-the-frontier
ShipTalk — https://www.shiptalk.io
Harness — https://www.harness.io
Follow
Bryan D. Payne: https://www.linkedin.com/in/bdpayne/
Adam Arellano: https://www.linkedin.com/in/adamrossarellano/   
Martin Reynolds: https://www.linkedin.com/in/martinreynolds/   
ShipTalk is brought to you by Harness. Subscribe wherever you listen, and until next time — let's stop talking and start shipping.

SpaceX bought Cursor's parent company Anysphere for $60 billion. OpenAI immediately cut off Cursor's access to its models; Anthropic went the other way and increased compute support. Business decision, or political one?
Adam Arellano and Martin Reynolds are joined by Papanii Okai, EVP of Product Engineering at Rocket and former CTO of Venmo, who opens with a warning that everyone is reading this deal wrong. His case: Cursor cracked the interface between humans and models, and that capability matters far more to a company building physical AI than any code editor. He also argues Apple — not OpenAI, not Microsoft — takes ambient AI, because the win condition isn't an app you launch.
The conversation moves to the talent math nobody wants to do. This year 32% of organizations skipped buying software and built it instead. Large enterprises doubled agent adoption while small firms stayed flat. Meanwhile US entry-level engineering postings fell 67%, and conventional wisdom says a senior takes 5 to 9 years to grow. Papanii thinks that timeline is about to compress hard — and that companies planning to stop hiring will hit a burnout wall instead of a scaling curve.
Also in this episode: whether Tim Cook's retirement signals a broader shift toward builder CEOs, why ransomware crews are choosing cheap repeatable playbooks over frontier models, and what still happens to a room of engineers when you say "Log4j" out loud.
ShipTalk is brought to you by Harness.

Key moments:
00:51 — SpaceX buys Cursor for $60B 
03:59 — Guest: Papanii Okai 
04:42 — The contrarian read on the deal 
06:23 — Why Apple wins ambient AI 
13:30 — Harness by the Numbers 
14:35 — Build vs. buy and the death of SaaS 
18:46 — Seniors are in their golden years 
20:00 — Where do 2030's seniors come from? 
25:43 — Do universities still matter? 
29:30 — Tim Cook retires, a builder takes over 
34:07 — The AI threat apocalypse isn't here 
39:22 — The one thing teams get wrong