Anthropic CODE RED, Claude Steals Your Memory, GPT-5.4 Released!

Collage just made it way easier to switch your AI provider. So, this is a smart move by Enthropic, and I want to show you what they just put out because it directly addresses one of the main reasons people don't switch AI tools, which is that if you spend months getting your current one to actually know you, like if you've been using Chat GPT for a while, you probably built up a bunch of contexts. It knows how you write, what kind of tone you like, what projects you're working on, what you do for work. That stuff accumulates over time and is genuinely useful. So the idea of starting fresh somewhere else and going through all of that again is just kind of annoying. It's a friction that keeps people from trying anything new. Enthropic basically said, "Okay, let's just remove that friction entirely. " They built this page at cloud. com/importmemory and the whole thing is just two steps. You go there, you copy a prompt they give you, you paste it into whatever AI you're currently using, chat GPT, Gemini, whatever, and that AI basically summarizes everything it knows about you, your preferences, your context, your working style, all of it. Then you take that output and then you paste it into Cloud's memory settings. That's it. Two, copy paste and Claude just inherits all of that context. The framing they use on this page is actually kind of good. They say your first conversation should feel like you're 100th, which is the actual problem they're solving. Most people who try a new AI tool end up leaving not because it wasn't good enough, but because the onboarding experiences are just exhausting. You have to reexlain everything and it feels like starting over. What's also worth noting is that Claude's memory system is set up to keep different projects and context separate so things don't bleed together and you can see and edit everything it remembers which is something people have been asking for across the board. That transparency of like what does this thing actually know about me and can I control it? The timing of this is pretty interesting too. With GPT 5. 4 just dropping and a lot of people evaluating what tools they're actually using dayto-day. Enthropic is essentially saying, "Okay, if you've been curious about cloud but haven't wanted to deal with the migration headache, here's the answer. Go to cloud. com/importmemory, spend about a minute on it, and just try it. " Whether that's enough to move people over is a different question, but as a product decision, it's pretty smart. The switching cost was a real barrier, and they just made it a lot

smaller. So, OpenAI just dropped GPT 5. 4, for and I want to talk about it because there's actually some stuff in here worth paying attention to. This isn't just them bumping a version number. The main thing they did is consolidate. For a while, they've had separate models optimized for different things. One that's better at coding, reasoning, and so on. GPT 5. 4 is basically them taking all of that and putting it into one model. So, you're not having to choose anymore. You just use the thing, and it handles both pretty well. The benchmark numbers they publish are pretty strong as well. There's an evaluation called GDP val that tests AI against actual professionals across 44 different occupations and GPT 5. 4 matched or beat those professionals 83% of the time. GBT 5. 2 was at 71% on the same test. So that's a real jump and not just a simple upgrade. They also added native computer use, meaning the model can actually sit down and operate a computer, click around, navigate apps, use a mouse and keyboard. That sounds like a small thing, but it's kind of a big deal for what people are building with these models right now. On the developer side, they added something called tool search. The old way of building on the API meant you had to manually pass in every possible tool your app might need every single time. Now the model just figures out what it needs, so it's cleaner to build with, and it reduces prompt size, which brings cost down. They also made it more efficient in general. It uses fewer tokens than 5. 2 to solve the same reasoning task, which again matters a lot if you're running this at scale. And AI hallucination, they say individual factual claims are 33% less likely to be wrong, and overall responses are 18% less likely to contain errors compared to 5. 2. Those numbers sound modest, but honestly, that's the kind of thing that makes a model actually usable for professional work rather than just impressive in demos. Availability wise, GBT 5. 4 thinking is rolling out now for plus, team, and pro users. The pro version is on pro and enterprise plans. API is $2. 50 per million input tokens, $12 per million output. The thing I keep coming back to with this release is just the pace. They dropped 5. 3 Instant on March 3rd and then this came out 2 days after. They're clearly in a sprint right now and GBT 5. 4 feels like the moment where a lot of the separate threads they've been pulling on, reasoning, coding, agenting stuff actually coming together in one coherent package. Whether that holds up outside of benchmarks, we'll see. But on paper, this is one of the more complete releases they put out in a while. Okay

so there's a lot happening with Enthropic right now. And I want to lay out the whole picture because it was genuinely one of the more wild stories intact recently. It started a few weeks ago. Enthropic had a $200 million contract with the Pentagon and it had two conditions baked in. Claude could not be used for fully autonomous weapons targeting and it cannot be used for mass domestic surveillance of Americans. Then the Pentagon under Pete Hagset came back and said, "We want to renegotiate everything under a standard any lawful use clause. " meaning no private company gets to put restrictions on what the military can do with the technology. Enthropic said no. The Pentagon gave them a hard deadline. Enthropic held firm and at 5:01 p. m. on a Friday, the Pentagon declared Enthropic a supply chain risk to national security. The first time that label has ever been applied to an American company. and Trump posted on Truth Social calling Enthropic a radical left woke company and ordered every federal agency to stop using their products. Within hours, OpenAI stepped in and signed the deal Anthropic had just refused. So, that's already a significant situation. But then a memo leaked. Dario Amodi had written a 1,600word internal memo to Anthropics employees on the same day everything fell apart and the information published it and it was not diplomatic. He called open AI's Pentagon deal maybe 20% real and 80% safety theater. Said Sam Elman's public statements were straight up lies and accused him of presenting himself as a peacemaker while signing a contract that undermined everything he publicly said he stood for. But the part that really blew things up was when Daria laid out why he thought the Trump administration came after Enthropic specifically. He wrote that the real reasons were that Enthropic hadn't donated to Trump while Open AI and Greg Brockman had donated a lot and they had given dictator style praise to Trump. Greg Brockman and his wife gave 25 million to a MAGA super PAC. Sam Elman donated a million to Trump's inauguration and stood next to him at the Stargate announcement. Dario did none of that and apparently believes that's a big part of why Enthropic is now on the wrong side of the administration. This memo leaked and it immediately complicated everything. A White House official told Axios it could blow up chances of resolution and the under secretary of defense went on to X to say there were no active negotiations with Enthropic, directly contradicting what Daario had publicly said about productive conversations happening. Daario has since apologized for the tone of the memo, saying it was written within hours of a chaotic series of announcements and didn't reflect his careful or considered views. But the damage is sort of already done. The Pentagon formally confirmed the supply chain designation and Anthropic says it plans to sue to challenge it. Here's where things get actually complicated for Anthropic as a business. They're reportedly preparing for an IPO, a supply chain risk designation, a lawsuit against the federal government, and a sitting president publicly calling your company radical and woke. That's a lot of baggage to carry into a public offering. Some of Enthropic's own investors are reportedly frustrated with one person briefed on the situation describing it as an ego and diplomacy problem. The counterpoint is that the public seems to have responded pretty well to anthropic stance. ChatGBT unstalled search after OpenAI's Pentagon deal and Claude climbed to number two in the app store. So in the court of public opinion, Daario may have actually won something. But in terms of federal contracts, government relationships, and what this means for the company's trajectory, that's a much messier picture and is still unfolding. But

that's it for today's video. Make sure to subscribe to the channel, follow us on Twitter, follow the world of AI, and don't forget to subscribe to our newsletter. We post constantly and you don't want to miss this.