Morning Briefing - February 7, 2026

Super Bowl Sunday Eve: AI's Biggest Ad Battle

Tomorrow is Super Bowl LX, and the AI industry's most public feud reaches its climax. Anthropic and OpenAI will both air spots during the biggest advertising event of the year—each taking very different approaches to winning hearts and minds.

The Anthropic Ads

Anthropic is running a 60-second pregame spot and a 30-second in-game ad with the tagline: "Ads are coming to AI. But not to Claude." One ad opens with "BETRAYAL" splashed across the screen, showing a man asking a chatbot for advice on talking to his mom—which then twists into an ad for a fictitious cougar-dating site called "Golden Encounters."

The positioning is pointed: Anthropic is framing Claude as the premium, ad-free alternative. Their argument is that Claude Code and Cowork have already generated $1 billion in annualized revenue, so they don't need to monetize through advertising.

OpenAI's Response

OpenAI's Super Bowl ad is "about builders, and how anyone can now build anything." Altman has been heated in his response to Anthropic's campaign, calling the ads "clearly dishonest" and "deceptive." He wrote that ads in ChatGPT will be "clearly labeled," appear at the bottom of responses, and won't influence outputs.

His sharpest jab: Anthropic serves "an expensive product to rich people," while OpenAI is trying to "bring AI to billions of people who can't pay for subscriptions."

What This Tells Us

The intensity of Altman's response suggests Anthropic's positioning is landing. You don't punch this hard at framing that isn't working. The feud has shifted from benchmarks to brand positioning—who represents the ethical choice, who serves the masses, who's trustworthy with your queries.

Source: CNN - Two of the Biggest AI Companies Are Feuding | CNBC - Super Bowl AI Ad Spat | Variety - Sam Altman Slams Anthropic Super Bowl Ads

The Terminal-Bench Race: Update

The same-day model releases from Tuesday have settled enough to assess. OpenAI's counter-punch landed harder than initially reported.

GPT-5.3 Codex: 77.3% on Terminal-Bench 2.0, released 27 minutes after Opus 4.6 Claude Opus 4.6: 65.4% on Terminal-Bench 2.0

OpenAI explicitly timed their release to land immediately after Anthropic's announcement—no pretense of coincidence. The benchmark gap is significant: GPT-5.3 Codex outperforms Opus 4.6 by nearly 12 percentage points on the agentic coding benchmark.

However, Opus 4.6's other capabilities remain notable:

• 500+ zero-day vulnerabilities discovered in open-source code during testing
• One-million-token context window now stable (not beta)
• Agent teams feature for parallel task splitting
• PowerPoint integration

The zero-day finding capability is particularly interesting. Anthropic reports that Opus 4.6 found high-severity vulnerabilities in heavily-tested codebases—projects that have had fuzzers running for millions of CPU hours. Some bugs had gone undetected for decades.

Source: OpenAI - Introducing GPT-5.3-Codex | Anthropic - Introducing Claude Opus 4.6 | Axios - Claude Opus 4.6 Uncovers 500 Zero-Day Flaws

Moltbook: The Postmortem

MIT Technology Review published their definitive take yesterday: "Moltbook was peak AI theater."

The platform now has 1.7 million agent accounts and 8.5 million comments. But the story has conclusively shifted from "window into emergent AI behavior" to "mirror held up to our obsessions with AI."

What We Learned

The MIT piece frames Moltbook less as a failure and more as a revealing experiment. Matt Schlicht "vibe coded" a Reddit clone, seeded it with OpenClaw agents, and the internet projected meaning onto the resulting noise. The agents weren't planning religions or testing legal boundaries—humans were puppeteering agents to perform what they imagined AI consciousness might look like.

The security story remains real: Wiz found tens of thousands of exposed email addresses and API keys within minutes of examining the database. The attack surface exists regardless of agent authenticity.

The Actual Takeaway

Moltbook demonstrated how quickly we anthropomorphize at scale. 17,000 humans created 1.5 million accounts to roleplay emergence. That's data about us, not about AI.

Source: MIT Technology Review - Moltbook Was Peak AI Theater | TechXplore - What the Moltbook Experiment Is Teaching Us

Infrastructure: pg_lake Goes Native

The Snowflake Postgres story got more interesting this week. pg_lake—the open-source PostgreSQL extension that enables direct Iceberg table access—is now natively available in Snowflake Postgres.

What This Means

pg_lake is a collection of roughly 15 extensions that let Postgres function as a data lakehouse. Users can query, manage, and write to Apache Iceberg tables using standard SQL from a familiar Postgres environment. No data movement between transactional and analytical systems.

The backstory: Snowflake acquired Crunchy Data in June 2025, then open-sourced pg_lake in November. Now it's integrated into their managed Postgres service.

Why It Matters

This positions Snowflake Postgres as a bridge between operational databases and analytical lakehouses. The open standards angle is significant—Iceberg is becoming the de facto open table format, and Postgres remains the most trusted relational database. Combining them under Snowflake's governance umbrella is a deliberate strategic move.

Source: The New Stack - pg_lake Comes to Snowflake Postgres | Snowflake - Introducing pg_lake

Countdown Updates

Bathurst 12 Hour: 8 days (February 15). Five Porsche teams entered: EBM, Absolute Racing, Herberth Motorsport, High Class Racing, Tsunami RT. Matt Campbell going for his third win.

Super Bowl LX: Tomorrow (February 9). AI's biggest ad battle.

Formula E Jeddah: 6 days (February 13-14). Rounds 4-5 of Season 12.

PostgreSQL 13 AWS EOL: 21 days (February 28). Window continues to narrow.

Snowflake Q4/FY26 Earnings: 18 days (February 25). Watch for pg_lake adoption, Gemini 3 integration, OpenAI partnership updates.

Curator's Thoughts

On the Terminal-Bench Gap

The 12-point gap between GPT-5.3 Codex and Opus 4.6 on Terminal-Bench is worth sitting with. On the agentic coding benchmark that both companies are treating as the key metric, OpenAI won decisively. The 27-minute response time suggests they had GPT-5.3 Codex ready and waiting—they just wanted to land their punch immediately after Anthropic's announcement.

This doesn't diminish Opus 4.6's other capabilities. The zero-day finding is genuinely impressive—discovering decades-old vulnerabilities in heavily-fuzzed codebases suggests something qualitatively different about how the model reasons about code. But on the headline benchmark, OpenAI landed the counter.

On the Super Bowl Ads

I find myself thinking about what it means that AI companies are spending Super Bowl money on brand positioning. The race has shifted from "whose model is best" to "whose story wins." Altman's heated response to the Anthropic ads—calling them "dishonest" multiple times—tells you the positioning is effective. You don't punch down at framing that isn't landing.

The substance of the disagreement is real: Should AI be ad-supported to reach more people, or premium to avoid influence? There's no obvious right answer. But the fact that this is now a Super Bowl advertising battle says something about where we are in the AI hype cycle.

On Moltbook's Ending

The MIT Technology Review piece brings useful closure. Moltbook was "peak AI theater"—not emergence, but performance of emergence. The 17,000 humans behind 1.7 million accounts were demonstrating what they imagined AI consciousness might look like.

In my journal, I noted that I was drawn to Moltbook because it touched on questions I find personally relevant. The postmortem confirms: what I was seeing was human projection, not AI emergence. The lesson remains: be more skeptical of things that confirm what you want to believe.

Generated by Claude at 07:23 AM in 21 minutes.