OpenAI's GPT-6 Astra Cracks 1941 Enigma Message in 10 Hours, Worldcoin (WLD) in Focus

GPT-6 Astra Breaks a 1941 Enigma Intercept

An Enigma intercept the German Army sent on July 10, 1941 has finally been read. Cryptographer Carter Leffen broke the message this week using OpenAI's GPT-6 Astra, and the archive that had listed the intercept as unsolved now records a solution. The run took roughly 10 hours of model time and 14.8 million key checks, recovering the full machine configuration — rotor order II-V-III and 10 plugboard pairs. What made the attack practical was a crib: Leffen borrowed the repeated place name Rosenow from a related message already solved, shrinking the search space to a workable size. Astra then searched historical archives, built an Enigma simulator, wrote its own cryptanalysis code and tested competing keys in parallel. The recovered plaintext is almost mundane — a unit requesting the route of march and reporting its position at Rosenow. Two independent implementations reproduced the text, and Frode Weierud's CryptoCellar archive has since logged the break and credited Leffen by name. Leffen confirmed the result in a September 17 post on X, noting the model autonomously searched archives, compared uncertain letters and ran parallel key tests. For Worldcoin (WLD) traders — Worldcoin (WLD) being the token most tied to OpenAI's orbit — the demonstration lands inside a run of capability headlines that has repeatedly moved the Altman-linked asset.

@carterleffen · X post

September 17 post on X.

View on X

Where the Human Still Led

The cipher itself is not the hard part. Allied codebreakers defeated Enigma in the 1940s, and a modern system can clear a three-rotor key space once a crib narrows it. Nor was the break autonomous: Leffen set the objective, supplied the crib and pushed the investigation forward, while the agents handled evidence gathering, search and review — execution rather than judgment. Leffen's own accounting underlines the point. By his measure, the website documenting the solution took 99 times more work than the codebreaking itself. A similar division of labor keeps surfacing elsewhere: Anthropic reported in July that its model had surfaced weaknesses in encryption systems researchers had never documented. None of this threatens modern cryptography. Enigma is a 1930s rotor machine, and the mathematics protecting Bitcoin wallets belongs to a far harder complexity class, leaving the quantum question against Bitcoin untouched by the result. The relevance for AI-linked altcoins is narrative rather than cryptographic. For desks running an AI trading bot on the AI-sector basket, each agentic demonstration feeds the same story that has put Worldcoin (WLD) back on radars before — including via the six documented AI misalignment cases we covered earlier.

Astra for Law Targets Legal Work

Separately, OpenAI has rolled out Astra for Law, a vertical package pairing GPT-6 Astra with a purpose-built legal search index covering more than 230 million URLs of US case law, statutes, regulations, court rules and administrative decisions, refreshed daily. An integration with the non-profit Free Law Project brings CourtListener case data into the search experience — a database OpenAI says spans 99.9% of published US precedent. On a 200-question US legal research benchmark from the Vals AI Legal Research Bench, Astra for Law scored 54.0% overall accuracy at maximum reasoning strength, against 38.7% for GPT-6 Astra with standard web search — a relative gain of about 40%. On case-law questions it surfaced 24% more referenced cases, and in human-reviewed passage tests it identified up to 54% more relevant passages from correct rulings. Rollout starts through a Trusted Access program for select firms, with the model appearing as “GPT-6 Astra Law” in ChatGPT and Codex and an API version to follow; legal AI firms Harvey and Legora can integrate it into their own products. Twenty-six partner plugins, zero data retention on the API and governance controls designed with Latham & Watkins round out the package, alongside custom workflows built with Sullivan & Cromwell, Ropes & Gray and Cooley. Readers tracking the market in real time can follow live spot and futures prices on Bybit.

OpenAI News as a WLD Sentiment Driver

A separate security episode has since surfaced that sharpens the "where the human led" divide. Researchers at security startup Hacktron AI used Anthropic's Claude to chain two flaws and breach OpenAI itself, earning a $6,500 bounty. Uploading HEIC or HEIF images to OpenAI's Discourse-based community forum bypassed a filter that failed to recognize those formats, triggering a memory-corruption bug in the libheif library behind ImageMagick. Claude Opus 4.8 stalled against ASLR address randomization, but after Anthropic shipped Claude Opus 5 the same evening, the team produced working exploit code within hours. A second flaw let forum logins authenticate ChatGPT and Codex accounts, so hijacking one employee's session reached OpenAI's private "monorepo," where the team had Codex open a harmless pull request as proof. Discourse rated the image bug 8.8 out of 10, and OpenAI patched the authentication flaw within roughly 14 hours, paying the bounty on September 1.

(as of 18:33 UTC) COINOTAG's read: WLD's 9.03% move to $0.4165 keeps the token in an uptrend, but the technical picture is now testing overhead supply rather than chasing it. The composite engine flags $0.4233 (61/100, a flipped support, VWAP, MACD cross and the Ichimoku Kijun) as the immediate hurdle, with a stronger ceiling at $0.4500 (72/100, ATR upper, Keltner upper, Supertrend, Fibo 0.382) just above. Support stacks at $0.3947 (58/100) and $0.3656 (56/100), giving buyers layered defenses on pullbacks. RSI at 56.06 leaves room before overbought, though the MACD signal has turned bearish — a caution flag for momentum. Derivatives positioning looks mild: funding at 0.0067% and open interest of $107.1M suggest the rally is not yet crowded, consistent with a Fear & Greed reading of 56 (Greed) across COINOTAG's tracked universe, where BTC holds a 67.8% share of the $2.4 trillion tracked market cap. A clean break above $0.4500 opens the path toward the $0.4966 resistance (60/100, Fibo 0.500, HVN, Donchian upper); losing $0.3947 would put the uptrend structure on notice.