OneHuman Watchdog: June 2026 — Five Tightenings, One Cut

In June 2026, five of the eight AI tools OneHuman tracks restricted, obscured, or unilaterally modified what their subscribers paid for. One moved the opposite direction. That asymmetry — five tightenings, one cut — is the structural observation this report records. It is the first time the pattern has been visible at the monthly level with enough documented events to name it plainly.

The tightening side shared a common property: in each case the subscriber was the last to know. Caps communicated through support emails. Model versions differing across surfaces of the same subscription. A government directive that pulled a product mid-subscription on day four. A class-action that put the question of whether the advertised multiplier matches delivered usage in front of a federal judge. The cutting side had competitive pressure and chose to respond by lowering prices rather than managing churn through opacity.

Both directions ran simultaneously, in the same 30 days. Subscribers who tracked both had leverage. Subscribers who did not noticed the throttles and missed the cut.

---

The Five Tightenings

Anthropic usage limits and throttle documentation — June 10. On June 3 and June 4, the same paying subscriber hit hard usage caps on two separate Anthropic products in 24 hours — Claude Cowork at 7:32 PM, Claude Chat at 10:47 AM the following morning. Anthropic had shipped a post-mortem on April 23 and reset all subscriber limits as compensation for three documented bugs. Six weeks later the throttle pattern continued. A second structural failure sat beneath the cap events: the same subscriber, on the same date and same subscription tier, saw Opus 4.8 in Claude Chat but only Opus 4.7 in Claude Cowork, twelve days after Opus 4.8 had launched. Differential model delivery across vendor surfaces, undisclosed. Full documentation in Anthropic Throttle.

AI tool naming as a category failure — June 14. Five naming decisions across four vendors in 60 days — each one making tier comparison structurally harder. The Anthropic Fable 5 case escalated the stakes. Anthropic launched Fable 5 on June 9 and described it as the most capable model it had ever made generally available. Alongside it, Mythos 5 — the same underlying model, safeguards lifted, deployed for approved government users. Subscribers saw "Fable" and could not know Mythos existed, that Fable was the safeguard-constrained variant, or that a jailbreak exploit was bypassing Fable's constraints to reach Mythos-level capability. On approximately June 11–12, the US Commerce Department issued an export-control directive; Anthropic could not reliably segment foreign-national traffic in real time and took both models offline globally. Subscribers who had upgraded got four days. The product is gone, refunds are the only remedy, no restoration timeline exists. A government-initiated mid-subscription product withdrawal — a category of subscriber risk that had not existed before. Full documentation in AI Tool Naming Opacity.

Google Gemini: three structural changes in 60 days — June 17. Between May 17 and June 8, Google introduced compute-based usage caps on paid Pro and Ultra tiers without subscriber notice, rebuilt its entire tier ladder at I/O 2026, and quietly cut AI Plus from $7.99 to $4.99 with no announcement. The caps were walked back after 11 days of community backlash — which itself confirms the introduction was silent, since you do not reverse what you disclosed clearly. The tier restructure was announced; the migration logic for existing subscribers was not published, and OneHuman could not find it. The price cut: existing subscribers were not told. They either discovered it or kept overpaying. OneHuman's automated 15-day pricing sweep caught it on June 15 — the vendor did not announce it. A March subscriber holds a materially different product by mid-June, through three separate unilateral modifications. Full documentation in Three Quiet Moves in 60 Days.

Anthropic Claude Max class-action — June 20. A federal class-action complaint filed June 15 in the U.S. District Court for the Northern District of California alleges Anthropic's Claude Max 20x plan ($200/month) delivers roughly six to eight times Pro usage, not the twenty the name implies. Max 5x ($100/month) is alleged to deliver about 3.5 times, not five. The plaintiff alleges a single five-hour session consumed 15 percent of his weekly Max 20x allowance. The allegations are unproven. The class is not certified. Anthropic declined to comment. The proceeding matters because the question of whether the advertised multiplier matches delivered usage is now in a venue that can compel disclosure — the public record this category has lacked. Full documentation in Anthropic Claude Max Class Action.

xAI SuperGrok Heavy throttle — June 25. SuperGrok Heavy's video cap fell from "over 100 videos per reset window" in April to "more than 80" in May, communicated through a support email with no blog post or changelog. The subscription page's language shifted from "Near-unlimited usage" to "Highest usage limits" — vague enough to stay technically accurate at any cap, without disclosing that the cap had moved. A public statement promised higher limits. By June, a third-party tracker cited a third number, different from both the support-ticket figure and the public promise. Failed generations still consume quota. The 720p allowance, once exhausted, silently drops to 480p with no notification. A discounted upgrade offer appears at the precise moment a subscriber's quota runs out. The pattern OneHuman is documenting is not that limits are wrong — compute costs money. It is that every number attached to this subscription has come from a tweet, a support ticket, or a tracker, never from xAI's own published documentation. Full documentation in xAI SuperGrok Throttle.

The sixth structural-opacity event — discovered June 29. On June 29, we documented a sixth structural-opacity event: Anthropic silently changed the Claude Code default model on April 30, 2026, in a way that compounds three cost multipliers (token price, tokenizer, default effort) into roughly a 2.4x per-call cost increase for unchanged workloads. The discovery emerged from investigating our own invoices — a recursive-opacity situation unique to OneHuman, which publishes via Claude Code. Extended treatment ships in a forthcoming BNA. The discovery is part of June's record because June is when it surfaced.

---

The Counter-Pattern

In the same 60-day window, DeepSeek moved the other direction.

On April 24, 2026, DeepSeek released V4-Pro and V4-Flash, both with a 1M-token context window. On May 22, it converted a 75% promotional discount on V4-Pro into the standing list price — permanently, not as a promotion. V4-Pro went from $1.74/$3.48 per million input/output tokens to $0.435/$0.87. V4-Flash is at $0.14/$0.28. The free consumer tier at chat.deepseek.com stayed free through the upgrade. OneHuman's June 15 pricing sweep recorded the same figures independently.

The story is not that DeepSeek is the answer. A watchdog that stops at the price cut and skips the trade-offs is doing marketing. The full set of trade-offs: data jurisdiction (DeepSeek's privacy policy acknowledges storage on Chinese-jurisdiction servers, where national law grants broad government access), active US federal restrictions (the Pentagon, NASA, and the Navy restrict the tool on government devices, with expansion legislation pending), open weights but closed training data and alignment (model weights are genuinely published; the layer that decides what the model will and won't say is not), pricing sustainability (a rate reported to run below market cost may be subsidized and a lower price today is not a guaranteed lower price next year), and CCP-aligned output moderation on politically sensitive queries.

The structural observation is narrower and more useful than "switch vendors." Competitive pressure, in the same 60-day window that subscriber lock-in produced caps discovered one support ticket at a time, produced a verified price cut and a capability upgrade. The leverage is real. Whether you pull it depends on your data. Full documentation in DeepSeek Counter-Pattern.

The asymmetry above — five tightenings, one cut — was documented by OneHuman's 15-day pricing sweep and continuous news monitoring, measured against vendor announcements that did not happen. The Gemini price cut on June 15 came from the sweep, not the vendor. The Anthropic throttle continued six weeks past the published post-mortem. The instruments OneHuman built to surface these patterns are why this report exists. They are not the story. The asymmetry is.

---

Verified by OneHuman · June 30, 2026

---

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.