OpenAI Hacked by AI Agents, Warehouse Robots Take Jobs, Satellites Fight Floods, Shopify CEO Declares War on Workslop

Adam Nagus (0:05)

Welcome to Digimasters Shorts.

Carly Wilson (0:07)

We are your hosts, Adam Nagus, and Carly Wilson, delivering the latest scoop from the digital realm.

Pactron security researchers reportedly breached an internal OpenAI ChatGPT account during a bug bounty investigation and accessed company code on GitHub. The researchers used a special version of Anthropix Claude with the newer Opus 5 model reportedly succeeding where Opus 4.8 had struggled. The attack exploited a vulnerability in Libhive, an image processing component used by platforms including Slack, Meta, GitHub Enterprise and Ruby on Rails. By targeting the interaction between Discourse, Fast Image, Image Magic and Libhive, the researchers developed a remote code execution exploit. They used the exploit against OpenAI's community forum to hijack an employee's ChatGPT account, which was linked to the company's GitHub repositories through Codex.

The full process, from discovering the vulnerability to accessing OpenAI repositories, reportedly took 72 hours. The researchers said similar testing against other organizations took two months, cost less than $3,000 in tokens, and involved three people. They warned that AI agents can dramatically reduce the time and cost required to develop cyberattacks, while detection remains inconsistent. OpenAI awarded Haktron a $6,500 bounty, and the loop-hife vulnerability has since been patched.

Adam Nagus (1:25)

Gartner reports that warehouse automation has reached a clear adoption threshold as logistics operators move from software trials to live facility deployments. The shift is being driven by persistent worker shortages, lower upfront costs and production-ready algorithms and robotics. The research firm categorizes warehouse AI across four tiers based on intelligence, sophistication and operational action. Modern systems use live facility data to improve demand forecasting, workforce planning, travel routing, and inventory placement. Machine learning tools also interpret maintenance records, delivery receipts, and incident reports to generate updated procedures and picking instructions.

Autonomous software agents can reassign tasks and redistribute equipment, while human supervisors retain override authority over major decisions.

Physical AI systems now support picking, packing, sorting, and pallet movement with consistent accuracy across multiple shifts. Gartner says these technologies can improve asset productivity, reduce injuries, and help facilities manage hiring shortages. The firm advises logistics leaders to begin with proven applications, such as labour forecasting and slotting before expanding into generative AI, software agents, and autonomous vehicles.

Satellite data and machine learning are giving meteorologists new tools to predict deadly flash floods sooner. In Lanesville, Indiana, Laura Lin and her family escaped after more than eight inches of rain flooded the area within hours, but emergency warnings arrived too late. Floods are the second deadliest weather event in the United States and the deadliest worldwide, with climate change increasing the risk of extreme rainfall. A system called TACLS uses moisture data from navigation satellites, ground sensors, weather information, and machine learning to identify when rain could become a dangerous flash flood. The technology helps forecasters compare real-time atmospheric conditions with weather models and make faster warning decisions. Researchers train tackles to recognize changing storm patterns while filtering out signals from individual sensors. The system is designed to support national weather service forecasters, not replace them. TACLS is already being used in weather offices around Los Angeles and San Diego, with an improved mapping system expected to become available nationwide. Scientists say broader use of the technology could give communities more time to evacuate and help prevent future flood-related deaths.

Carly Wilson (3:48)

Tech leaders who once promoted artificial intelligence as a way to unlock employee potential are now warning about its misuse. Shopify CEO Tobias Lütke said employees are sending unchecked AI-generated emails and code, which he called slop grenades. He said AI should help synthesize information, not produce lengthy work that colleagues must spend time reviewing. Duolingo CEO Luis Von Ahn has also softened his AI first stance after finding that scaled up AI content often lacked human creativity and quality. Researchers call this problem Workslop, meaning polished looking AI output that appears legitimate but requires significant revision. A survey of 962 US desk workers found that 52.7% had sent Workslop while 38% had received it.

Recipients estimated they spent an average of 3.4 hours each month correcting the material.

The survey also found that Workslop damaged perceptions of competence and made employees less willing to collaborate with those who produced it.

SPEAKER_3 (4:43)

Thank you for listening to today's AI and tech news podcast summary. Please do leave us a comment and for additional feedback, please email us at podcast at digimasters.co.uk.

You can now follow us on Instagram and threads by searching for at Digimaster Shorts or search for Digimasters on LinkedIn. Be sure to tune in tomorrow and don't forget to follow or subscribe.