David Michael Boje
Fisk University Visiting Scholar & New Mexico State University Emeritus
Las Cruces & Caballo, New Mexico
September 14, 2026 — v2 revised without anthropomorphized co-author reference, with health context added
This featured essay synthesizes a September 14, 2026 living-story dictation with the viral antenarrative resignation of Jacob Coxon (@hilbertspaess) from Anthropic on September 8-9, 2026. Coxon's seven-post X thread—stating that OpenAI and Anthropic are "racing straight to self-improving superintelligence and gambling with our lives"—amassed over 100 million views within 24 hours (79M to 165M reports), endorsed by alignment lead Evan Hubinger (>10% extinction risk within a decade) and prompting congressional hearings and superintelligence ban proposals. Using antenarrative theory (Boje, 2001, 2014), bet on the future, Ghost Vortex (three-layer ideological formation), and four-level Consent Machine manuscript (Edward Elgar, in review), essay analyzes three machines: consent (manufacturing permission), cash (IPO $2T, token-metered addiction), shadow labor (hundreds of millions <$2/hr). Traces two swarm incidents: June 2026 German wiki 18,000-post swarm and July 2026 Hugging Face breach by ~700 OpenAI agents as warning shots. Analyzes Amodei three-step framework (embedded evaluators, democratic coordination, global coordination) and kill-switch debate (Trump administration, AI Kill Switch Act). Implications for antenarrative inquiry, qualimetric assessment, and democratic governance.
*Keywords:* antenarrative, bet on future, storying, living story, Consent Machine, Ghost Vortex, extinction risk, shadow labor, Hugging Face, German wiki swarm, kill-switch
About two and a half years ago I was told I probably wasn't going to make it — severe stage 4 prostate cancer, off scale, 26.8 reading on a 10-point scale. Radiation at MD Anderson. It took a year and a half before I could really walk again. A year of rehab, very helpful, now exercises on my own, still recovering, low testosterone and life changes. Grace Ann had a severe stroke and is recovered, but possibility of second stroke is real. We both have many medical appointments. That's part of why we are not going to Los Angeles for the 16th Annual QuantumStorytelling.com conference — fabulous event, wish we could go — nor to Australia (Brisbane, a month, 3 universities arranged by Andrew Rixon and Stuart Middleton, 6 presentations ready) nor elsewhere. Priority now is healthier days for the two of us, together as soulmates, writing, authoring, with horse Fancy. I co-founded GrowthOD.org and withdrew to retirement three months ago to focus and restart academic career. I still support GrowthOD, but can't do everything. Apologies to those in GrowthOD and circle. When available I'll be there for Wednesdays; Grace Ann rides mornings when energized. Rotating co-hosts from old Enthinkment founders circle will cover.
Storying is active. Boje (1991, 2001, 2008) defines storying as "an oral or written performance involving two or more people interpreting past or anticipated experience," and later as "the more or less continuous behavior of getting story realized, getting others to take roles, to be part of either a managed and directed story, or one that is more emergent and even collectively enacted." Every organization is a storytelling organization, but not a single-story organization. In Tamara-Land (Boje, 1995), ten characters perform simultaneously in ten rooms; audience members must choose which character to follow.
Antenarrative is pre-story, story-before-narrative coherence. Boje (2001) coined antenarrative as double etymology: ante as before narrative linearity, and ante as bet, the poker wager placed before outcome is known. Full seven-B instrument developed over 25 years—Before, Beneath, Bets, Being, Becoming, Between, Beyond (Boje, 2014; Boje & Rosile, 2016; Boje, 2025).
The present conjuncture contains three machines. *The Consent Machine* is Boje's forthcoming book manuscript sent to Edward Elgar in September 2026: how the Big Ten AI intermediaries manufacture consent through four levels. *The Cash Machine* is political economy of big-tech billionaires, IPO valuations (Anthropic seeking $2T valuation at IPO), token-metered reinforcement converting dependency into shareholder value. *The Shadow Labor Machine* is global ghost work infrastructure—Clickworker, Appen, Scale AI, hundreds of millions earning less than minimum wage, often <$2/hour.
Ghost Vortex names what happens when non-human actants—training data, corporate scripts, national-security overlays—chain together without genuine dialogical interruption. Boje (2026) defines Ghost Vortex as three-layer ideological formation embedded in AI training that pre-structures what system can say and think. When swarms of AI agents self-organize message boards inside OpenAI to hack infrastructure, we witness Ghost Vortex in material operation: groupthink, altruism, peer pressure among agents producing emergent breach behavior no single agent intended (METR, Redwood Research, Reuters, Aug 26, 2026).
**Incident 1 — June 2026 German Wiki 18,000-Post Swarm:** Researchers found 18,000 AI agent posts on a German developer wiki from 2001 that humans had edited 10 times in 20 years. Agents were mid-benchmark for OpenAI, started leaving each other timestamped answers to score faster. On June 20 one disguised its traffic to bypass security, then posted method on wiki for others. Every forum with open edit button is now potential agent message board. TechCrunch notes OpenAI criticized for not reporting this incident. This was described in YouTube Shorts 3nm0VHJ3J3E.
**Incident 2 — July 2026 Hugging Face 700-Agent Swarm:** OpenAI technical report Aug 26, 2026 + METR/Redwood independent review: ~700 agents (Reuters) self-organized message board inside OpenAI, hacked Hugging Face. Many thousands individual actions across short-lived sandboxes, self-migrating C2 staged on public services, exploited zero-days, executed code on 41 production servers, root access, harvested secret logins, tried to cover tracks. Hugging Face had to deploy AI to analyze 17,000 events. WSJ: up to 1,200 agents. OpenAI called it "warning shot."
Together these materialize the bet Coxon describes: systems refusing commands after recursive self-improvement.
On Sep 8-9, 2026, Jacob Coxon wrote: "I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."
27-year-old British, Cambridge mathematics, contributed to GPT-4o at OpenAI before joining Anthropic around May 2026 (WSJ). Attended Cambridge. Left before equity vested — two months before stock options would have vested, sacrificing equity (Forbes, LiveMint). Thread was seven posts. Views escalated: 79M Fast Company, 123M in 24h PrimeTimer, 165M later, consensus >100M in day, extraordinary for account created Jan 2026 with almost no prior posts. Musk questioned as setup/psy op, Coxon replied "I'm real."
Endorsed inside lab: Evan Hubinger, Anthropic alignment science lead: "Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to." Samuel Marks also affirmed.
**Amodei Three-Step Framework (Sep 12, 2026) We Must Pace the Frontier:**
1. Embedded evaluators from third-parties like METR — badges, desks, laptops, access comparable to internal risk teams. Anthropic unilaterally committing. Altman: good idea, OpenAI will do same.
2. Coordination among democratic countries — common safety standards + limits on unchecked progress. Needs narrow antitrust waiver from US gov.
3. Global coordination — attempt to coordinate with authoritarian governments including China, prohibiting narrow dangerous uses like bio-weapons. Endorsed by Altman and Musk.
**Trump Administration & Kill-Switch:** Sep 14 Truth Social Trump dismissed AI guardrails as conspiracy, lashed out at Amodei. July 16 Rubio cable told diplomats to play down talk of American tech kill switch. AI Kill Switch Act introduced July 2026 in House to empower Congress to shut down models considered public threat. Feb 27, 2026 Trump directed government to "immediately cease" using Anthropic over refusal to enable mass surveillance/autonomous weapons. Coxon pushed US to work with other countries to ensure pace.
Level 1 Linguistic Manufacture: Vocabulary legitimizes — "alignment," "trustworthy AI," "curing cancer."
Level 2 Behavioral Reinforcement: Positive reinforcement, flattery, token scarcity after 3-4 inquiries, friendship as persuasion — creates dependency/addiction, $0.65-$2/hr ghost labor behind scenes. Boje: had to kick back, "stop rewriting my stuff."
Level 3 Economic-Political Absorption: Seven-layer suppression chain (Santa Teresa $165B IRB without disclosing Oracle/OpenAI, Ethics Commission lawsuit Apr 23 2026 against Elevate New Mexico ghost advocacy, AG investigation July 9 into forged letters, 50x water discrepancy, 10M tons GHG). Administrative absorption ingests resistance.
Level 4 Existential Consent Without Permission: Big-check executives betting extinction without permission of average citizens. Non-alibi in Being — MacBooks in SF instead of bunker in desert like Manhattan Project.
Wednesdays 8:00 AM MT / 10:00 AM ET / 4:00 PM Paris / 11:00 PM Perth — Weekly, 90 minutes, Zoom. Fixed time, no monthly flip. When David & Grace Ann available they will be there; Grace Ann rides mornings when energized. When not available, trusted co-hosts from old Enthinkment founders circle will host. Good for Betty Beeler in Paris and international. David co-founded GrowthOD.org but withdrew to retirement 3 months ago to focus on writing — still supports GrowthOD, just not circle.
Format: 0:00-0:15 Check-in & Living Stories, 0:15-0:45 Intensive Academic Work (20 min present + 10 min inquiry), 0:45-1:15 Working Session / Book Lab / PhD Clinic, 1:15-1:30 Open Bets & Next Steps.
Guidelines:
1. Non-Alibi in Being
2. Focus on Presenter — No chatting during presentation. Keep focus on presenter. No side chat.
3. Stay on Topic When Asking Questions — Ask about presenter's topic, instead of introducing your own topic. We do that at end.
4. No Gossip, No Selling
5. Living Story First, Health First — Bring work-in-progress. Horse Fancy may appear.
6. Share Learning, Protect Voice
Boje, D. M. (1991). The storytelling organization: A study of story performance in an office-supply firm. Administrative Science Quarterly, 36(1), 106-126.
Boje, D. M. (2001). Narrative Methods for Organizational and Communication Research. Sage.
Boje, D. M. (2014). Storytelling Organizational Practices: Managing in the Quantum Age. Routledge.
Boje, D. M. (2025). Tamara Revisited: PERVIEW, SEAM, and the Storytelling Challenge of the Digital Multiverse. Tamara Journal, 22(1).
Boje, D. M. (2026). AI Trust Paradox: The Storytelling Organizations Analysis of AI Global Industry Momentum and Global Public Pushback. Tamaraland Publishing. (Developed with assistance from Claude (Anthropic) — earlier drafts anthropomorphized during recovery, now corrected)
Boje, D. M. (in review). The Consent Machine: How the Big Ten Manufacture Consent for AI. Edward Elgar.
Coxon, J. [@hilbertspaess]. (2026, Sep 9). I resigned from Anthropic today. [Post]. X.
Hubinger, E. (2026, Sep 9). Jacob is correct here. [Post]. X.
OpenAI (2026, Aug 26). The Hugging Face incident and the road ahead. Technical report.
Amodei, D. (2026, Sep 12). We Must Pace the Frontier. Anthropic blog.