# ChatGPT Lets You Rewrite a GIF One Frame at a Time

Canonical URL: https://superpowerdaily.com/videos/0850b3b3-4421-4157-8150-22d3bd8f201f
Published: 2026-11-05T17:48:47.000Z
Episode date: 2026-11-05
Language: English
Duration: 8:29
Publisher: Superpower Daily
YouTube: https://www.youtube.com/watch?v=ylZCpNtZOv4

OpenAI's Images 2.6 brings editable, frame-by-frame GIFs to ChatGPT, as voters push back on data-center plans and Anthropic says Claude helped map a viral protein shell whose biological purpose remains unknown. OpenAI also says a prototype generated Linux privilege-escalation exploits, but they require an existing low-privilege foothold and fixes still need human review.

## Key takeaways

- Editable flipbook-style GIFs offer frame-level control, but独
- Election wins strengthen AI critics without immediately en
- Claude accelerated protein research, but protocol errors s
- AI scales exploit discovery, while human maintainers must

## Chapters

- 0:34 Editable GIFs: https://superpowerdaily.com/videos/0850b3b3-4421-4157-8150-22d3bd8f201f?t=34
- 2:33 Election Pushback: https://superpowerdaily.com/videos/0850b3b3-4421-4157-8150-22d3bd8f201f?t=153
- 4:11 Protein Mapping: https://superpowerdaily.com/videos/0850b3b3-4421-4157-8150-22d3bd8f201f?t=251
- 5:46 Linux Security: https://superpowerdaily.com/videos/0850b3b3-4421-4157-8150-22d3bd8f201f?t=346

## Stories covered

- OpenAI adds editable GIF creation to ChatGPT with its Images 2.6 release: https://superpowerdaily.com/posts/openai-adds-editable-gif-creation-to-chatgpt-with-its-images-2-6-release
  The release pairs short animations with new editing controls and developer models aimed at speed or visual consistency. Its performance gains remain company claims.
- Abdul El-Sayed wins Michigan Senate seat after campaigning for tighter AI controls: https://superpowerdaily.com/posts/abdul-el-sayed-wins-michigan-senate-seat-after-campaigning-for-tighter-ai-controls
  Victories for candidates seeking tighter AI controls strengthen the case for scrutiny. Congressional investigations are a forecast, not an announced next step.
- Anthropic uses Claude to map a 96-protein vault from a bacteria-infecting virus: https://superpowerdaily.com/posts/anthropic-uses-claude-to-map-a-96-protein-vault-from-a-bacteria-infecting-virus
  The study gives researchers a simpler system for investigating mysterious protein shells. Claude helped throughout, but one protocol error cost two weeks.
- OpenAI says its AI pipeline built 948 exploits that give Linux users root access: https://superpowerdaily.com/posts/openai-says-its-ai-pipeline-built-948-exploits-that-give-linux-users-root-access
  The prototype paired attack demonstrations with candidate fixes. Its results put the spotlight on validating and prioritizing bugs, not just discovering them.

## Transcript

Automatically transcribed from the narration.

[0:00] Hey everyone Let's jump right into this Welcome to the Superpower Daily AI Briefing for November 5th 2026 Today we're unpacking four massive market and research stories that are literally reshaping how we build with govern and interact with artificial intelligence We've got a lot of ground to cover in today's explainer so let's get right to it Here is our roadmap for today First OpenAI Images 2.6 Second Election Wins for Scrutiny Third Claude Maps a Protein Vault And

[0:28] fourth OpenAI's Linux Kernel Exploits Then we'll wrap it all up with four key takeaways Starting with Section 1 OpenAI Images 2.6 Editable GIFs hit ChatGPT and the API So if you've ever tried to generate a short animation and got frustrated because your character's face completely morphed halfway through this is for you ChatGPT can now take your text prompt a photo or even just a simple sketch and turn it into an editable GIF Now what's really fascinating

[0:57] is how it does this under the hood It uses a flipbook animation mechanism Rather than generating a single video file from scratch the model actually generates a series of related still images and then strings them together Why does this matter Well it gives you unprecedented control Inside ChatGPT you can now inspect that sequence frame by frame tweak the playback speed and crucially revise one individual frame before you share the final animation It puts the control right

[1:24] back in your editing workflow And for the developers out there using the API OpenAI is taking a brilliant new dual approach They've split the API into two completely distinct models On one side we have Flare Flare is built for pure speed and low complexity We're talking about churning out a low quality image in roughly six seconds On the other side we have Sunburst Sunburst is heavily optimized for high quality and complex scenes focusing on preserving visual

[1:50] details across multiple edits The API also drops some amazing new developer tools You can now mark specific regions for edits separate scene planning complexity from the actual image quality and even export your designs as scalable SVG files For web and logo design that SVG feature is absolutely huge But of course we have to look at the catch OpenAI is claiming these speeds are 50% faster than their older models Keep in mind though those are internal company

[2:18] claims When OrcaRouter checked on November 5th neither Flare nor Sunburst had actually appeared on the artificial analysis leaderboards yet So the key open question is whether those blazing fast speeds hold up in broader independent testing Moving on to section two election wins for scrutiny Voters and candidates push back on AI infrastructure Let's look at how the massive physical footprint of AI is playing out in the political arena Across recent elections candidates who campaigned for tighter AI

[2:48] controls like Abdul El Sayed winning a Senate seat in Michigan secured some very significant victories And the scale of this localized pushback is incredible According to Wired there were 40 local data center ballot questions across 10 different states including 21 in Ohio alone Voters are directly weighing in on development plans and rezoning NBC News even found that nearly a quarter of voters consider data centers a crucial factor in their vote Naturally this grassroots momentum faced heavy

[3:17] industry pushback In South Carolina for instance a pro industry super PAC dropped $1.5 million backing candidate Darlene Graham Despite that spending she still lost to an opponent who favored a data center moratorium And this sentiment crosses party lines too We're seeing incoming governors in both Georgia and Texas backing data center moratoriums Now it's important to be clear here This doesn't mean we're about to see a nationwide pause on AI The practical immediate change is simply a

[3:46] lot more political scrutiny and risk around local construction And as for federal action don't expect any instant laws The Washington Post expects Congress to use its oversight powers first conducting investigations and examining the industry before they even try to draft actual AI legislation So while advocates for tighter rules definitely have a stronger voice now formal federal oversight is going to take time All right section three Claude maps protein fault Anthropic resolves a 96 protein shell from

[4:18] a virus We are shifting gears entirely to a staggering biological breakthrough Anthropic researchers working with their Claude science model have just mapped the very first 3D structure of a vault particle from a bacteria infecting virus We're looking at a hollow shell built from 96 identical copies of a protein called L05 And how AI accelerated this workflow is just amazing Step one Claude searched through 1.5 billion predicted proteins Step two researchers actually produced those L05 protein candidates

[4:49] in leaving E coli bacteria Step three they used cryo EM imaging which essentially flash freezes the particles to confirm the 3D vault shapes By duplicating segments they even figured out how to make the shell 20% larger which is huge for studying how these structures assemble Why does this matter One word speed Claude suggested a totally new preparation method When the team tested it it worked flawlessly cutting preparation time down from three days to just 90 minutes

[5:18] Then running on hundreds of GPUs in parallel they had a full first model ready in just 34 hours But here is the reality check AI still makes mistakes One growth protocol that Claude drafted actually included a bad additive That single mistake cost the team two whole weeks of delay Furthermore while we now have this gorgeous 3D map of a 96 protein shell its actual biological function inside the cell remains a total mystery Finally section four OpenAI's

[5:49] Linux kernel exploits An AI pipeline moves from finding bugs to demonstrating them We're looking at a wild intersection of AI and cybersecurity here OpenAI built a prototype security pipeline that autonomously audits the Linux kernel and the output from just a 12 hour run is mind blowing The AI agents audited 2.8 million lines of code That's about 7% of the entire kernel They produced over 2,300 candidate findings But most impressively the pipeline actively generated 948 local privilege

[6:20] escalation exploits on stock virtual machines It didn't just find the bugs it built the exploits to prove them But this kind of compute isn't cheap This single audit run cost a staggering $1.2 million specifically $1,213,430 in API credits Let's define local privilege escalation really so we don't panic This means the AI figured out how to take an existing low privilege foothold on a machine and turn it into complete root access It does not mean the AI

[6:52] is remotely breaking into computers over the internet And while the AI is incredibly fast at generating these exploits and even suggesting patches human maintainers still have to review and validate every fix So while AI completely scales discovery the actual mitigation still requires human expertise Section five four key takeaways Synthesizing today's AI market movements We've covered a lot today Here is your synthesis Number one images 2.6 brings editable flipbook style GIFs to chat GPT offering flair for

[7:25] speed and sunburst for quality though independent benchmarks are still pending Number two the political landscape for AI is tightening locally with huge scrutiny on data centers but federal laws are going to wait on investigations first Number three Anthropx Cloud radically speeds up biological prep for things like mapping a 96 protein virus shell even though costly AI errors prove we absolutely still need human experts And number four OpenAI's security pipeline proved AI can autonomously build 948 local

[7:54] Linux exploits in hours meaning the bottleneck in cybersecurity is shifting from finding bugs to the human cost of fixing them All of this points to a massive theme We're seeing AI slash biological prep time to minutes and turn out hundreds of cyber exploits in hours So the question I want to leave you with today is this As AI accelerates everything from protein mapping to cyber exploits are we moving faster than our oversight can follow Think about

[8:21] that as you watch these markets evolve Read more at superpowerdaily com Stay curious

## Editorial standards

https://superpowerdaily.com/editorial-standards

This episode reports information available at publication time. Consult the linked stories and their reporting sources for context and subsequent updates.