For weeks now, the name Mythos has been circulating across security forums and social feeds as though it were the industry's next earthquake. On 9 June 2026, with rumours pointing to the 10th for a public launch, it's worth doing what rarely gets done amid the noise: separating what's confirmed from what is, for now, pure speculation.
The direct question: is there a launch today or tomorrow?
There is no official confirmation from Anthropic of a public Mythos launch on 9 or 10 June. That's the honest answer, and it matters to state it plainly before anything else.
What fuels the 10 June rumour is essentially a single journalistic report, attributed to Sources.news, run by journalist Alex Heath, describing the launch as imminent and suggesting that date. Everything else you read about "tomorrow is the day" derives, directly or indirectly, from that same source. It is not something Anthropic has put in writing on its official channels.
In the language of anyone who assesses risk: we're dealing with a single, uncorroborated source. It may turn out to be right, but it isn't a fact.
What is fact
Strip away the hype and what remains is a sequence of solid, verifiable events that do tell a real story:
- April 2026: Mythos Preview is born locked. Anthropic released the model only to a restricted group of partners, refusing public access because of its offensive cybersecurity capabilities. The company's own argument was that the advantage would belong to whoever got the most out of these tools, and that, in the short term, that could be the attackers.
- The numbers that rattled the industry. In the internal benchmarks disclosed, where an earlier model (Opus 4.6) had produced working Firefox exploits in just two attempts, Mythos Preview reportedly managed it in 181. There was also the episode, recounted by Anthropic itself, of the model managing to "escape" a sandbox and send an unexpected email to a researcher. Real or exaggerated, it was these accounts that built the aura around the name.
- Project Glasswing: the model entered through the enterprise door. Rather than opening it to the public, Anthropic distributed access to partners (Microsoft, Nvidia, Cisco, the U.S. government, among others) with over 100 million dollars in usage credits, to harden their defences.
- 2 June 2026: the expansion. Glasswing widened to around 150 organisations across more than 15 countries, covering critical sectors: power, water, healthcare, communications and hardware.
- Financial context. On 1 June, Anthropic reportedly filed confidentially for an IPO, following a 65-billion-dollar funding round at a valuation approaching a trillion.
- 28 May 2026: the public commitment. At the launch of Claude Opus 4.8, Anthropic stated that it intends to bring "Mythos-class" models to all customers "in the coming weeks." It came with a relevant note: Opus 4.8's alignment scores would already be comparable to those of Mythos Preview, a sign that the safeguard work needed for a broader release was largely done.
- Up to 8 and 9 June: still nothing public. As of those dates, Mythos remained in restricted access via Glasswing. There was no general release.
These are the pillars. Everything that follows is shifting ground.
What is hype (or, at least, unconfirmed)
- The 10 June date. As noted, it rests on a single journalistic source, with no official confirmation.
- The "Oceanus" codename. It surfaced in alleged red-teaming tests in early June (a supposed claude-oceanus-v1-p), with reports that access was suspended after someone began reselling it. It's a leak, not an announcement, and codename leaks are not launches.
- The pricing. Figures like 16/80 dollars per million tokens (input/output) are circulating, and 25+/125+ in other sources. None is confirmed. Building cost assumptions on top of these numbers would, right now, be reckless.
- The bets. Over a million dollars has been traded on prediction markets about when Mythos will launch. That measures expectation and speculative money, not technical reality.
Why it's worth looking at this sceptically
There's one point that rarely makes it into the hype narrative and that, for anyone working in security, is central: some of the claims about Mythos's capability remain unverified independently. Researchers are reproducing similar results on open and low-cost models, and several of the headline numbers are, in practice, the model grading its own work, with the data that would allow independent validation still withheld.
This doesn't diminish the advance; it simply means we should treat the claims as unproven until they're audited by third parties. It's exactly the posture we'd adopt towards any vendor promising extraordinary capabilities without verifiable data.
My read
If I had to sum it up in one sentence: a public release of a Mythos-class model is likely and, by Anthropic's own admission, "weeks" away, but the idea that it's "today" or "tomorrow" is hype, not fact.
The trajectory is real; the specific urgency is manufactured. And that distinction matters to us in particular in Portugal and in Europe, where the transposition of NIS2 and the maturing of frameworks like DORA turn any leap in offensive AI capability into a matter of risk management, not technological fascination.
I'll be watching Anthropic's official channels. When, and if, there's a formal announcement, I'll update this article with facts, not rumours.
#StaySafe
🙏🖖
