Skip to content
The Crash Log
AI & Tech Gone Off the Rails
Fund
Cover image for The Crash Log newsletter
Issue #020 · May 19, 2026

The Default Install

Four signals from the week AI moved from optional to operating system.

FATAL_EXCEPTION

Figure's Humanoid Robots Cross 100,000 Packages in 5-Day Nonstop Warehouse Sort

Figure AI began streaming what it called an eight-hour endurance test of its Figure 03 humanoid robots on May 13, with three of the units sorting barcoded packages onto a warehouse conveyor belt. By Monday the livestream still had not stopped. As of this writing, the fleet has been running for more than 100 hours of continuous autonomous operation, has sorted upward of 100,000 packages, and is holding roughly 1,200 to 1,250 packages per hour at a 2.9-second cadence. (Source: TechRadar).

The benchmark Figure chose to publicize was three seconds per package — parity with a fast human sorter. The variable the demo erased was the rest of the shift. Human sorters typically process 300 to 600 items per hour and lose throughput to fatigue, breaks, shift changes, and the injury rates that warehouse operations budgets exist to contain. Running on a single Helix-02 vision model with no operator intervention, Figure's units kept the rhythm through the human sleep cycle (Source: Interesting Engineering).

The demonstration did not name the customer. The arithmetic is now in front of every logistics buyer. Warehouse sorting and material-handling work is paid at or near the U.S. labor-force floor and concentrates disproportionately among Latino and immigrant workers — pickers and movers turn over at high rates, unionize at low ones, and have absorbed the brunt of every prior wave of warehouse automation.

The Figure 03 fleet does not turn over. The procurement question is no longer whether humanoid sorting is technically possible. It is which vendor wins the contract (Source: BigGo Finance).

STACK_OVERFLOW

OpenAI Turns Malta Into First Country to Default to ChatGPT

OpenAI and the government of Malta announced a national partnership this week that will give every Maltese citizen one free year of ChatGPT Plus after completing a government-sponsored "AI for All" online course developed by the University of Malta. The program is being administered by the Malta Digital Innovation Authority. Economy Minister Silvio Schembri called the agreement "the first partnership of such scale" (Source: Euronews).

Malta's population is roughly 600,000 — small enough that the per-citizen cost to OpenAI is trivial, but large enough to credibly call it a national rollout. The official framing is education and digital literacy. The mechanic is something else: the first sovereign state to install ChatGPT as the default tool that its teachers, workers, students, and ministries learn to think through. Once a country's professional habits form inside one model's interface, switching stops being a product choice and becomes an infrastructure migration (Source: The Next Web).

The Malta deal lands the same week the U.S. Center for AI Standards and Innovation finalized pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI, and as the European Union continues pressing Anthropic for access to its Mythos cybersecurity model. National governments are now negotiating with frontier labs at two layers at once: who gets to test the models before release, and whose citizens get to live inside them (Source: CNBC).

ACCESS_DENIED

OpenAI Adds Plaid Integration, Connects ChatGPT to Bank Accounts Across 12,000 Institutions

OpenAI launched a personal-finance preview inside ChatGPT on May 15 that lets users connect their bank, brokerage, and credit-card accounts through Plaid. The integration supports more than 12,000 financial institutions, including Chase, Schwab, Fidelity, Robinhood, American Express, and Capital One. Once linked, ChatGPT renders a dashboard of spending, portfolio performance, subscriptions, and upcoming payments, and accepts natural-language queries against all of it (Source: TechCrunch).

The preview is read-only at launch — ChatGPT cannot move money, file taxes, or trade. OpenAI says Intuit support is planned next, which will route tax estimates, credit-card approval odds, and live-clinician handoff through the same interface. Plaid's involvement is the operational shoulder of the deal: the same pipes that let Venmo and Robinhood read account balances now feed the largest consumer chatbot in the world (Source: Quartz).

The risk has moved. The launch is positioned as a passive read view, but Intuit integration and the broader OpenAI agent roadmap make clear the next step is to extend reads into writes (Source: Engadget).

Stack Trace

Cisco notified roughly 4,000 employees on May 14 that their roles were eliminated — the same day the company reported $15.8 billion in quarterly revenue, up 12 percent year over year. Executives framed the cuts as cost-structure reallocation toward silicon, optics, security, and enterprise AI. Severance and related charges will run to roughly $1 billion. This is Cisco's second 4,000-headcount round in just over a year (Source: TechCrunch).

Source: TechCrunch

OpenAI agreed this week to grant the European Union pre-deployment access to its new cyber model. Anthropic, more than a month after releasing Mythos, has still not granted EU regulators preview access — even as the model continues to be restricted to a small set of cleared U.S. enterprise customers. The asymmetry between the two leading labs on European cooperation is now itself the news (Source: CNBC).

Source: CNBC

On May 12, X user @SHL0MS posted one of Claude Monet's Water Lilies paintings, labeled it as having been "Made with AI," and asked for replies to describe what made it inferior to a real Monet. Hundreds of users dutifully critiqued composition, depth, reflections, and "emotionless" colors — on a Monet. The experiment was a stunt, but it produced a clean data point on how the AI label now precedes the work in many viewers' judgment (Source: PetaPixel).

Source: PetaPixel

Don't miss the next issue

Subscribe