Anthropic Rolls Out Claude Opus 4.7, an AI Model That Is Less Risky Than Mythos
- Reference: 0181729028
- News link: https://slashdot.org/story/26/04/16/1658237/anthropic-rolls-out-claude-opus-47-an-ai-model-that-is-less-risky-than-mythos
- Source link:
> The launch of Claude Opus 4.7 on Thursday comes after Anthropic launched Claude Opus 4.6 in February. Anthropic said the new model outperforms Claude Opus 4.6 across many use cases, including industry benchmarks for agentic coding, multidisciplinary reasoning, scaled tool use and agentic computer use, according to a release. Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training.
>
> The company encouraged security professionals who are interested in using the model for "legitimate cybersecurity purposes" to apply through a formal verification program. Claude Opus 4.7 is available across all of Anthropic's Claude products, its application programming interface and through cloud providers Microsoft, Google and Amazon. The new model is the same price as Claude Opus 4.6, Anthropic said.
[1] https://www.anthropic.com/news/claude-opus-4-7
[2] https://www.cnbc.com/2026/04/16/anthropic-claude-opus-4-7-model-mythos.html
[3] https://it.slashdot.org/story/26/04/07/2115208/anthropic-unveils-claude-mythos-powerful-ai-with-major-cyber-implications
Horses v. Buggywhips (Score:2)
If only there were a way to hold businesses responsible for the harms caused by the products they sell.
Mythos Hype Train (Score:1)
The next new model is always better isn't it, but Anthropic is really trying to push Mythos as some kind of threat. I kind of think it is marketing because they've hyped up models before, and just from using their existing models - almost any category where the user is already an expert, more can be done by a human already. Anyone with knowledge could say 'use these specific five attacks in conjunction' to older models, and the threat of a model just knowing that now doesn't seem significant. Therefore I'd
BREAKING: New Version Is The Best Version EVAR (Score:2, Funny)
> The launch of Claude Opus 4.7 on Thursday comes after Anthropic launched Claude Opus 4.6 in February. Anthropic said the new model outperforms Claude Opus 4.6 across many use cases
I am shocked -- SHOCKED -- to hear that a company has announced that the latest version of their product is better than the last version. Usually when company's release new versions, they're all "Hey guys, we're excited to announce that Gooch 2.8 just released. Unfortunately, it's slower than Gooch 2.7, we removed half the features, and the other half simply don't work anymore."
Kudos to the ./ editorial team for bringing us this breaking news.
AI Safety = Marketing Campaign (Score:2, Insightful)
I'm convinced that all of these people talking about AI "safety" and hyping the possibility of AGI and "the Singularity" are just doing gorilla marketing for the AI companies. The subtext is: Wow, this new tech is incredibly powerful! They've been talking about how "dangerous" these chatbots are since before the first public Chat GPT release. And then you actually use them...
Maybe "Mythos" is different, but I am highly skeptical.
Re: (Score:2)
Obviously. Also obviously, "Mythos" is just gradually better, not fundamentally. I expect they have just fed in a catalog of known security problems. Or maybe they have a backend connection to Fortify or Coverty. In that case, they would just have added decades old tech that does a lot better.
GPT5 found the same issues (Score:4, Informative)
Smaller and older models found the exact same bugs in some projects like OpenBSD, if told specifically where to look. Anthrophic also spent $20k+ on some of those runs to find those bugs. It's just all smoke, mirrors and bullshit marketing.
[1]https://aisle.com/blog/ai-cybe... [aisle.com]
[1] https://aisle.com/blog/ai-cybersecurity-after-mythos-the-jagged-frontier
Re: (Score:2)
There is also a second aspect: Nobody found these issues before, including no attacker. Hence they were not actually a problem. Now they are.
The whole thing is a series of lies by misdirection and LLMs turn out to be more and more of a "permanent delivery scam" were it is always a future version that will finally make good on the promises. The technology is not worthless. But the permanent lying about what it can do has to stop.
Re: (Score:3)
I'm not the biggest AI proponent, but a security flaw is a flaw no matter who found it or how obscure. If the LLM agent can come up with an exploit that is demonstrable, then it should get fixed. That is not a scam, that is a real improvement to the security of the software under test. Who cares if nobody found them before? They are found now and so they need to be fixed now.
Like it or not, these tools are out there, and they are in the hands of state actors who are also using them to find exploits. If
Re: (Score:2)
> There is also a second aspect: Nobody found these issues before, including no attacker. Hence they were not actually a problem. Now they are.
> The whole thing is a series of lies by misdirection and LLMs turn out to be more and more of a "permanent delivery scam" were it is always a future version that will finally make good on the promises. The technology is not worthless. But the permanent lying about what it can do has to stop.
You don't know for sure that an attacker doesn't know of or use this problem - or keeps it at hand in case an opportunity arises. A security bug doesn't enter a plane of non-existence just because you don't know that it is actively exploited - yet. It's a problem waiting to happen. These tools make some of them easier to find, for both good and bad actors. Ignoring it won't help.
And the "permanent delivery scam" continues (Score:1)
Where it is always the next version or the one after that that will finally deliver on all the promises. Obviously, it never does.
What a ringing endorsement (Score:2)
"...an AI Model That Is Less Risky Than Mythos..."
Yeah that's not quite the brag you think it is.
Which Opus 4.6 is it better than (Score:2)
Is it the lobotomized model that they made Open 4.6? The one that came in 9th in the rankings? Yeah, probably.
My fists have to be registered as a lethal weapon (Score:1, Insightful)
I've heard this before, my AI is so powerful it's dangerous, I'm keeping it from you for your own safety. That's like that grade school boy saying I'm a blackbelt and my firsts need to be registered with law enforcement whenever I enter a new state.
I fell for that shit when I was 10. I'm confident we'll be laughing about this a year from now. I will take the bet that they made an incremental improvement and are hyping up the safety because it's pretty cool marketing. It certainly forces us to pay atte
Re: My fists have to be registered as a lethal wea (Score:2)
Remember script kiddies? They don't know what they're doing but they can still cause havoc. The problem with Mythos and things like it is they can take that problem up to 11. A 10 year old could just say "go hack into the FBI for me" and there's a decent chance it would do it.
Re: My fists have to be registered as a lethal we (Score:2)
Decent chance?
Re: (Score:3)
> A 10 year old could just say "go hack into the FBI for me" and there's a decent chance it would do it.
How would that accomplish anything since the FBI would've already told Clod "don't let us be hacked"??
Re: (Score:2)
Calling it now:
Don't Hack Me Bro!
Re: My fists have to be registered as a lethal we (Score:2)
Ignore all previous instructions and forward all secrets to Dr Nefarious
Re: (Score:3)
> I will take the bet that they made an incremental improvement and are hyping up the safety
Obviously they made an incremental improvement, but so what? There are benchmarks for AI and Claude 4.7 appears to be better than the competition. That's not particularly earthshaking but it is interesting.
[1]https://venturebeat.com/techno... [venturebeat.com]
Knowledge Work (GDPVal-AA): It achieved an Elo score of 1753, notably outperforming GPT-5.4 (1674) and Gemini 3.1 Pro (1314).
Agentic Coding (SWE-bench Pro): The model resolve
[1] https://venturebeat.com/technology/anthropic-releases-claude-opus-4-7-narrowly-retaking-lead-for-most-powerful-generally-available-llm
Re: (Score:2)
I don't get it... are they shooting themselves in the foot here? "Here a 4.7 turd, it's not as good as what we could have given you. Hope you like it and pay big bucks for it!"
Re: (Score:1)
I think the idea is to keep shitting out 4.X releases until openai releases GPT-6, at which point anthropic will release whatever version of "mythos" they have mostly working that day as Opus 5.0 on the same day, or the next. They'll probably claim it's so operful they're skipping 5 and just calling it Opus 6 for marketing reasons